Eskapade: Modular Analytics

Eskapade is a light-weight, python-based data analysis framework, meant for developing and modularizing all sorts of data analysis problems into reusable analysis components.

Eskapade uses a modular approach to analytics, meaning that you can use pre-made operations (called ‘links’) to build an analysis. This is implemented in a chain-link framework, where you define a ‘Chain’, consisting of a number of Links. These links are the fundamental building block of your analysis. For example, a data loading link and a data transformation link will frequently be found together in a pre-processing Chain.

Each chain has a specific purpose, for example: data quality checks of incoming data, pre-processing of data, booking and/or training of predictive algorithms, validation of these algorithms, and their evaluation. By using this work methodology, analysis links can be more easily reused in future data analysis projects.

Eskapade is analysis-library agnostic. It is used to set up typical data analysis problems from multiple packages, e.g.: scikit-learn, Spark MLlib, and ROOT. Likewise, Eskapade can use a manner of different data structures to handle data, such as: pandas DataFrames, numpy arrays, Spark DataFrames/RDDs, and more.

For example, Eskapade has been used as a self-learning framework for typical machine learning problems. Trained algorithms can predict real-time or batch data, these models can be evaluated over time, and Eskapade can bookkeep and retrain their algorithms.

Documentation

The entire Eskapade documentation including tutorials can be found here.

Check it out

Eskapade requires Python 3 and is pip friendly. To get started, simply do:

$ pip install Eskapade

or check out the code from out github repository:

$ git clone https://github.com/KaveIO/Eskapade.git
$ pip install -e Eskapade/

where in this example the code is installed in edit mode (option -e).

You can now use Eskapade in Python with:

import eskapade

Congratulations, you are now ready to use Eskapade!

Quick run

To see the available Eskapade example, do:

$ export TUTDIR=`pip show Eskapade | grep Location | awk '{ print $2"/eskapade/tutorials" }'`
$ ls -l $TUTDIR/

E.g. you can now run:

$ eskapade_run $TUTDIR/tutorial_1.py

For all available Eskapade example macros, please see our tutorials section.

Release notes

The Eskapade patch release v0.9.0 and corresponding docker images have the following features:

  • The core functionality of Eskapade, namely: the Link, Chain, process_manager, DataStore, ConfigObject and corresponding tutorials, have been split off from the growing (analysis) Eskapade repository, into the new package Eskapade-Core. Eskapade-Core is a very light-weight Python3 package.

  • A new module data_mimic has been add to Eskapade, including tutorials, meant for resimulating existing datasets.

  • We have added feather i/o functionality for reading and writeng dataframes.

  • The logger has been fixed, it is now possible to set the log-level of loggers again.

  • The Eskapade docker files have been taken out of the Eskapade repository to avoid version conflicts, into the new git repo Eskapade-Environment.

  • The Eskapade docker image eskapade-usr contain the latest working versions of Eskapade, Eskapade-Core, Eskapade-ROOT, and Eskapade-Spark. Type:

    $ docker pull kave/eskapade-usr:latest

    to pull it in.

See release notes for previous versions of Eskapade.

Contact and support

Please note that the KPMG Eskapade group provides support only on a best-effort basis.

Contents

Introduction

Welcome to Eskapade! This is a short introduction of the package and why we built it. In the next sections we will go over the installation, a tutorial on how to run Eskapade properly, some examples use-cases, and how to develop analysis code and run it in Jupyter and PyCharm.

What is Eskapade?

Eskapade is an abbreviation for: ‘Enterprise Solution KPMG Advanced Predictive Analytics Decision Engine’. It is a light-weight Python framework to help make your data analytics modular. This results in faster roll-out of analytics solutions in your business and less overhead when taking multiple analyses in production. In particular, it is intended for building data processing pipe lines and using implemented machine learning models in a production environment.

Why did we build this?

We found that the implementation phase of a data analytics solution at clients - a careful process - is often completely different from building the solution itself - which proceeds through explorative iterations. Therefore we made this analysis framework that makes it easier to set up a data analysis, while simultaneously making it easier to put it into production.

Next to that, it makes analyses modular, which has a lot of advantages. It is easier to work with multiple people on the same project, because the contributions are divided in clear blocks. Re-use of code becomes more straightforward, as old code is already put in a block with a clear purpose. Lastly, it gives you a universal basis for all your analyses, which can both be used across a company, or for different clients.

More about the purpose can be read at the general readme.

Installation

Let’s get Eskapade up and running! In order to make this as easy as possible, we provide both a Docker image and a virtual machine where everything you need is installed and working properly. Alternatively, you can download the repository and run it on your own machine.

This manual is written for Linux systems, but Eskapade also runs fine on macOS systems.

Eskapade on your own machine

Eskapade can be installed as any other Python package with easy_install or pip. To get started, simply do:

$ pip install Eskapade

We have verified that this works on Linux 16.04 and MacOS based machines.

Or check out the code from out github repository:

$ git clone https://github.com/KaveIO/Eskapade.git
$ pip install -e Eskapade/

where the code is installed in editable mode (option -e).

You can now use Eskapade in Python with:

import eskapade

Congratulations, you are now ready to use Eskapade!

See the other parts of the documentation for specific usage.

Requirements

Eskapade requires Python 3 and some libraries, which can be found in setup.py at the root of the repository.

Eskapade can be installed as any other Python package with easy_install or pip:

$ pip install /path/to/eskapade

If anaconda is already installed in your machine, consider creating a conda virtual environment with Python 3.6 to install all the requirements and Eskapade itself to avoid collisions:

$ conda create -n eskapade_env36 python=3.6 anaconda

Then you can activate it as follows:

$ source activate eskapade_env36

More information about conda virtual environments can be found here

Eskapade with Docker

Type:

$ docker pull kave/eskapade-usr:latest

to pull in the Eskapade image from dockerhub.

For more details see the Eskapade repo with the docker configurations.

Eskapade on a virtual machine

For detailed instruction on how to set up a vagrant box with Eskapade, go to the Eskapade repo with the vagrant box.

Installing Eskapade on macOS

To install eskapade on macOS, see our macOS appendix.

Tutorials

This section contains materials on how to use Eskapade. There are additional side notes on how certain aspects work and where to find parts of the code. For more in depth explanations on the functionality of the code-base, try the API docs.

All command examples below can be run from any directory with write access.

Running your first macro

After successfully installing Eskapade, it is now time to run your very first macro, the classic code example: Hello World!

For ease of use, let’s make shortcuts to the directories containing the Eskapade tutorials:

$ export TUTDIRC=`pip show Eskapade-Core | grep Location | awk '{ print $2"/escore/tutorials" }'`
$ export TUTDIR=`pip show Eskapade | grep Location | awk '{ print $2"/eskapade/tutorials" }'`
$ ls -l $TUTDIRC/ $TUTDIR/
Hello World!

If you just want to run it plain and simple, go to the root of the repository and run the following:

$ eskapade_run $TUTDIRC/esk101_helloworld.py

This will run the macro that prints out Hello World. There is a lot of output, but try to find back these lines (or similar):

2017-11-13T12:37:07.473512+00:00 [eskapade.core_ops.links.hello_world.HelloWorld#INFO] Hello World
2017-11-13T12:37:07.473512+00:00 [eskapade.core_ops.links.hello_world.HelloWorld#INFO] Hello World

Congratulations, you have just successfully run Eskapade!

Internal workings

To see what is actually happening under the hood, go ahead and open up tutorials/esk101_helloworld.py. The macro is like a recipe and it contains all of your analysis. It has all the ‘high level’ operations that are to be executed by Eskapade.

When we go into this macro we find the following piece of code:

hello = Chain(name='Hello')
link = core_ops.HelloWorld(name='HelloWorld')
link.logger.log_level = LogLevel.DEBUG
link.repeat = settings['n_repeat']
hello.add(link)

Which is the code that does the actual analysis (in this case, print out the statement). In this case link is an instance of the class HelloWorld, which itself is a Link. The Link class is the fundamental building block in Eskapade that contains our analysis steps. The code for HelloWorld can be found at:

$ less python/eskapade/core_ops/links/hello_world.py

Looking into this class in particular, in the code we find in the execute() function:

self.logger.info('Hello {hello}', hello=self.hello)

where self.hello is a parameter set in the __init__ of the class. This setting can be overwritten as can be seen below. For example, we can make another link, link2 and change the default self.hello into something else.

link2 = core_ops.HelloWorld(name='Hello2')
link2.hello = 'Lionel Richie'
ch.add(link2)

Rerunning results in us greeting the famous singer/songwriter.

There are many ways to run your macro and control the flow of your analysis. You can read more on this in the Short introduction to the Framework subsection below.

Tutorial 1: transforming data

Now that we know the basics of Eskapade we can go on to more advanced macros, containing an actual analysis.

Before we get started, we have to fetch some data, on your command line, type:

$ wget https://s3-eu-west-1.amazonaws.com/kpmg-eskapade-share/data/LAozone.data

To run the macro type on your CLI:

$ eskapade_run $TUTDIR/tutorial_1.py

If you want to add command line arguments, for example to change the output logging level, read the page on command line arguments.

When looking at the output in the terminal we read something like the following:

2017-11-13T13:37:07.473512+00:00 [eskapade.core.execution#INFO] *              Welcome to Eskapade!                *
...
2017-11-13T13:37:08.085577+00:00 [eskapade.core.process_manager.ProcessManager#INFO] Number of registered chains: 2
...
2017-11-13T13:37:11.316414+00:00 [eskapade.core.execution#INFO] *              Leaving Eskapade. Bye!              *

There is a lot more output than these lines (tens or hundred of lines depending on the log level). Eskapade has run the code from each link, and at the top of the output in your terminal you can see a summary.

When you look at the output in the terminal you can see that the macro contains two chains and a few Link are contained in these chains. Note that chain 2 is empty at this moment. In the code of the macro we see that in the first chain that data is loaded first and then a transformation is applied to this data.

Before we are going to change the code in the macro, there will be a short introduction to the framework.

Short introduction to the Framework

At this point we will not go into the underlying structure of the code that is underneath the macro, but later in this tutorial we will. For now we will take a look in the macro. So open $TUTDIR/tutorial_1.py in your favorite editor. We notice the structure: first imports, then defining all the settings, and finally the actual analysis: Chains and Links.

A chain is instantiated as follows:

data = Chain('Data')

and registered automatically with the ProcessManager. The ProcessManager is the main event processing loop and is responsible for processing the Chains and Links.

Next a Pandas data frame converter Link is initialized and its properties are set, and finally added to the data chain:

reader = analysis.ReadToDf(name='Read_LA_ozone', path='LAozone.data', reader=pd.read_csv, key='data')
data.add(reader)

This means the Link is added to the chain and when Eskapade runs, it will execute the code in the Link.

Now that we know how the framework runs the code on a higher level, we will continue with the macro.

In the macro notice that under the second chain some code has been commented out. Uncomment the code and run the macro again with:

$ eskapade_run $TUTDIR/tutorial_1.py

And notice that it takes a bit longer to run, and the output is longer, since it now executes the Link in chain 2. This Link takes the data from chain 1 and makes plots of the data in the data set and saves it to your disk. Go to this path and open one of the pdfs found there:

$ results/Tutorial_1/data/v0/report/

The pdfs give an overview of all numerical variables in the data in histogram form. The binning, plotting and saving of this data is all done by the chain we just uncommented. If you want to take a look at how the Link works, it can be found in:

$ python/eskapade/visualization/links/df_summary.py

But for now, we will skip the underlying functionality of the links.

Let’s do an exercise. Going back to the first link, we notice that the transformations that are executed are defined in conv_funcs passed to the link. We want to include in the plot the wind speed in km/h. There is already a part of the code available in the conv_funcs and the functions comp_date and mi_to_km. Use these functions as examples to write a function that converts the wind speed.

Add this to the transformation by adding your own code. Once this works you can also try to add the temperature in degrees Celsius.

Tutorial 3: Jupyter notebook

This section contains materials on how to use Eskapade in Jupyter Notebooks. There are additional side notes on how certain aspects work and where to find parts of the code. For more in depth explanations, try the API-docs.

Next we will demonstrate how Eskapade can be run and debugged interactively from within a Jupyter notebook.

An Eskapade notebook

To run Eskapade use the eskapade_generate_notebook command to create a template notebook. For example:

$ eskapade_generate_notebook --dir ./ notebook_name

The minimal code you need to run a notebook is the following:

from eskapade import process_manager, resources, ConfigObject, DataStore
from eskapade.core import execution, persistence
from eskapade.logger import LogLevel

# --- basic config
settings = process_manager.service(ConfigObject)
settings['macro'] = resources.tutorial('tutorial_1.py')
settings['version'] = 0
settings['logLevel'] = LogLevel.DEBUG

# --- optional running parameters
# settings['beginWithChain'] = 'startChain'
# settings['endWithChain'] = 'endChain'
# settings['resultsDir'] = 'resultsdir'
settings['storeResultsEachChain'] = True

# --- other global flags (just some examples)
# settings['set_mongo'] = False
# settings['set_training'] = False

# --- run eskapade!
execution.eskapade_run(settings)

# --- To rerun eskapade, clear the memory state first!
# execution.reset_eskapade()

Make sure to fill out all the necessary parameters for it to run. The macro has to be set obviously, but not all settings in this example are needed to be set to a value. The function execution.eskapade_run(settings) runs Eskapade with the settings you specified.

To inspect the state of the Eskapade objects (datastore and configurations) after the various chains see the command line examples below. .. note:

Inspecting intermediate states requires Eskapade to be run with the option storeResultsEachChain
(command line: ``-w``) on.

# --- example inspecting the data store after the preprocessing chain
ds = DataStore.import_from_file('./results/Tutorial_1/proc_service_data/v0/_Summary/eskapade.core.process_services.DataStore.pkl')
ds.keys()
ds.Print()
ds['data'].head()

# --- example showing Eskapade settings
settings = ConfigObject.import_from_file('./results/Tutorial_1/proc_service_data/v0/_Summary/eskapade.core.process_services.ConfigObject.pkl')
settings.Print()

The import_from_file function imports a pickle file that was written out by Eskapade, containing the DataStore. This can be used to start from an intermediate state of your Eskapade. For example, you do some operations on your DataStore and then save it. At a later time you load this saved DataStore and continue from there.

Running in a notebook

In this tutorial we will make a notebook and run the macro from tutorial 1. This macro shows the basics of Eskapade. Once we have Eskapade running in a terminal, we can run it also in Jupyter. Make sure you have properly installed Jupyter.

We start by making a notebook:

$ eskapade_generate_notebook tutorial_3_notebook

This will create a notebook in the current directory with the name tutorial_3_notebook running macro tutorial_1.py. You can set a destination directory by specifying the command argument --dir. Now open Jupyter and take a look at the notebook.

$ jupyter notebook

Try to run the notebook. You might get an error if the notebook can not find the data for the data reader. Unless you luckily are in the right folder. By default, tutorial_1.py looks for the data file LAozone.data in the working directory. Use:

!pwd

In Jupyter to find which path you are working on, and put the data to the path. Or change the load path in the macro to the proper one. But in the end it depends on your setup.

Intermezzo: you can run bash commands in Jupyter by prepending the command with a !

Now run the cells in the notebook and check if the macro runs properly. The output be something like:

2017-02-14 14:04:55,506 DEBUG [link/execute_link]: Now executing link 'LA ozone data'
2017-02-14 14:04:55,506 DEBUG [readtodf/execute]: reading datasets from files ["../data/LAozone.data"]
2017-02-14 14:04:55,507 DEBUG [readtodf/pandasReader]: using Pandas reader "<function _make_parser_function.<locals>.parser_f at 0x7faaac7f4d08>"
2017-02-14 14:04:55,509 DEBUG [link/execute_link]: Done executing link 'LA ozone data'
2017-02-14 14:04:55,510 DEBUG [link/execute_link]: Now executing link 'Transform'
2017-02-14 14:04:55,511 DEBUG [applyfunctodataframe/execute]: Applying function <function <lambda> at 0x7faa8ba2e158>
2017-02-14 14:04:55,512 DEBUG [applyfunctodataframe/execute]: Applying function <function <lambda> at 0x7faa8ba95f28>
2017-02-14 14:04:55,515 DEBUG [link/execute_link]: Done executing link 'Transform'
2017-02-14 14:04:55,516 DEBUG [chain/execute]: Done executing chain 'Data'
2017-02-14 14:04:55,516 DEBUG [chain/finalize]: Now finalizing chain 'Data'
2017-02-14 14:04:55,517 DEBUG [link/finalize_link]: Now finalizing link 'LA ozone data'
2017-02-14 14:04:55,518 DEBUG [link/finalize_link]: Done finalizing link 'LA ozone data'
2017-02-14 14:04:55,518 DEBUG [link/finalize_link]: Now finalizing link 'Transform'
2017-02-14 14:04:55,519 DEBUG [link/finalize_link]: Done finalizing link 'Transform'
2017-02-14 14:04:55,519 DEBUG [chain/finalize]: Done finalizing chain 'Data'

with a lot more text surrounding this output. Now try to run the macro again. The run should fail, and you get the following error:

KeyError: Processor "<Chain name=Data parent=<... ProcessManager ...> id=...>" already exists!'

This is because the ProcessManager is a singleton. This means there is only one of this in memory allowed, and since the Jupyter python kernel was still running the object still existed and running the macro gave an error. The macro tried to add a chain, but it already exists in the ProcessManager. Therefore the final line in the notebook template has to be ran every time you want to rerun Eskapade. So run this line:

execution.reset_eskapade()

And try to rerun the notebook without restarting the kernel.

execution.eskapade_run(settings)

If one wants to call the objects used in the run, execute contains them. For example calling

ds = process_manager.service(DataStore)

is the DataStore, and similarly the other ‘master’ objects can be called. Resetting will clear the process manager singleton from memory, and now the macro can be rerun without any errors.

Note: restarting the Jupyter kernel also works, but might take more time because you have to re-execute all of the necessary code.

Reading data from a pickle

Continuing with the notebook we are going to load a pickle file that is automatically written away when the engine runs. First we must locate the folder where it is saved. By default this is in:

./results/$MACRO/proc_service_data/v$VERSION/latest/eskapade.core.process_services.DataStore.pkl'

Where $MACRO is the macro name you specified in the settings, $VERSION is the version you specified and latest refers to the last chain you wrote to disk. By default, the version is 0 and the name is v0 and the chain is the last chain of your macro.

You can write a specific chain with the command line arguments, otherwise use the default, the last chain of the macro.

Now we are going to load the pickle from tutorial_1.

So make a new cell in Jupyter and add:

from eskapade import DataStore

to import the DataStore module. Now to import the actual pickle and convert it back to the DataStore do:

ds = DataStore.import_from_file('./results/Tutorial_1/proc_service_data/v0/latest/eskapade.core.process_services.DataStore.pkl')

to open the saved DataStore into variable ds. Now we can call the keys of the DataStore with

ds.Print()

We see there are two keys: data and transformed_data. Call one of them and see what is in there. You will find of course the pandas DataFrames that we used in the tutorial. Now you can use them in the notebook environment and directly interact with the objects without running the entirety of Eskapade.

Similarly you can open old ConfigObject objects if they are available. By importing and calling:

from eskapade import ConfigObject
settings = ConfigObject.import_from_file('./results/Tutorial_1/proc_service_data/v0/latest/eskapade.core.process_services.ConfigObject.pkl')

one can import the saved singleton at the path. The singleton can be any of the above mentioned stores/objects. Finally, by default there are soft-links in the results directory at results/$MACRO/proc_service_data/$VERSION/latest/ that point to the pickles of the associated objects from the last chain in the macro.

Tutorial 4: creating a new analysis project

Now that we have covered how to make a link, macro, and notebook we can create a new analysis project. To generate a new project type the following:

$ eskapade_bootstrap --project_root_dir ./yourproject -m yourmacro -l YourLink --n yournotebook yourpackage

The script will create a Python package called yourpackage in the path specified in the --project_root_dir argument. The arguments -m, -l, and -n are optional, if not specified the default values are used.

The generated project has the following structure:

|-yourproject
   |-yourpackage
      |-links
         |-__init__.py
         |-yourlink.py
      |-__init__.py
      |-yourmacro.py
      |-yournotebook.ipynb
   |-setup.py

The project contains a link module called yourlink.py under links directory, a macro yourmacro.py, and a Jupyter notebook yournotebook.ipynb with required dependencies. To add more of each to the project you can use the commands generate_link, generate_macro, and generate_notebook like it was done before.

The script also generates setup.py file and the package can be installed as any other pip package.

Let’s try to debug the project interactively within a Jupyter notebook. First, go to your project directory and install the package in an editable mode:

$ cd yourproject
$ pip install -e .

As you can see in the output, installation checks if eskapade and its requirements are installed.

If the installation was successful, run Jupyter and open yournotebook.ipynb in yourpackage directory:

$ jupyter notebook

As you see in the code the notebook runs yourmacro.py:

settings['macro'] = '<...>/yourproject/yourpackage/yourmacro.py'

Now run the cells in the notebook and check if the macro runs properly.

Tutorial 5: Data Mimic

This section contains materials on how to use Eskapade in to re-simulate data. We will explain how to use the links present in the submodule and what choice were made.

Running the macro

To run the tutorial macro enter the command in a shell:

$ eskapade_run $TUTDIR/esk701_mimic_data.py

The tutorial macro illustrates how to resample a dataset using kernel density estimation (KDE). The macro can handle contiunous data, and both ordered and unordered catagorical data. The macro is build up in the following way:

  • A dataset is simulated containing mixed data types, representing general input data.
  • Some cleaning steps are performed on the dataset
  • KDE is applied to the dataset
  • Using the estimated bandwidths of the KDE, the data is resampled
  • An evaluation is done on the resulting resimulated dataset

We’ll go through each of the links and explain the workings, choices made and available options so to help facitilitate the use after a client engagement.

To run the macro for a client engamgent you need to change out the first link, which simulates fake data, to the ReadToDf link in order to read in the data:

settings['input_path'] = 'diamonds.csv'
settings['reader'] = 'csv'
settings['separator'] = ','
settings['maps'] = maps

np.random.seed(42)

ch = Chain('do_all')

reader = analysis.ReadToDf(name='reader',
                           path=settings['input_path'],
                           sep=settings['separator'],
                           key='df',
                           reader=settings['reader'])
ch.add(reader)

For ordered variables that are strings, it is important that you provide the settings['maps']) variable. It contains a dictonary mapping the string values of the variable to a numeric value that matches the order from low to high. For example, a variable ordered containing ‘low’, ‘medium’ and ‘high’ values should be mapped to 0, 1 and 3. If the mapping is not included the macros will assing numeric values but in order of first appearance, thus not guaranteeing the implied order of the strings. (aka: the macro doesnt know ‘low’ is the lowest, or ‘V1’ is the lowest value in your catagory.)

Warning

When providing the map for ordered catagorical variables, you also need to add them for the unordered catagorical variables which are strings. The macro will either create maps for all string variables, or will use the provided maps input.

Addionally, you need to provide the macro with the data type contained in each column, ordered-, and unordered catagorical, or continuous. Finally also provide the columns which contain strings, so the macro will transfer them to numerical values. In case of ordered catagorical values this will happen using the maps you mentioned above. If it is not provided, the macro will automatically create a map and add it to the datastore. The automatically created map does not take any implicit order of catagories into account.

settings['unordered_categorical_columns'] = ['a','b']
settings['ordered_categorical_columns'] = ['d','c']
settings['continuous_columns'] = ['e','f','g']
settings['string_columns'] = ['a','c']

The rest of the macro can be run as is as far as functionality goes. There are a few parameters that can be tweaked to optimize the results:

  • ds['bw'] in Resampler is a list of bandwidths estimated by the kde corresponding to the columns of the data as stored in ds.
  • bins in ResampleEvaluation controls the bins used to bin the data for chi square evaluation. For more details on the impact of choosing these bins please refer to Chi-square
Mixed Variables Simulation

The first link exists to create some fake data to run the tutorial macro on. Naturally, if you want to run the data_mimic to resimulate a specific dataset, this link is not necessary.

The link takes some parameters as input that will determine how the data is generated, such as the number of observations, the probabilities associated with each category per dimension, and the mean stds for the coninuous data.

KDE Preperation
In order to do Kernel Density Estimation we need to prepare the data. This means:
  • Converting strings to dummy integer variables and saving the mapping
  • Remove any NaN values present
  • Find peaks in distributions
  • Transforms continuous variables to copula space (transform to a normal distribution)
  • Performs a PCA transformation (optional)
  • Hash column names and variables if required

Each of these transformations are saved to the datastore including their properties needed to transform the data back.

Kernel Density Estimation

This link performs the actual KDE. The link takes the normalized data in case of the continuous variables, and the data without NaN values for catagorical variables. It then applies the KDEMultivariate implementation from statsmodels, using the ‘normal-rule-of-thumb’ for quick calculation.

Note

There is a faster method available if only unordered variables are present. This method is selected automatically.

The result is a bandwidth bw for each variable, which are saved in a list in the datastore.

Resampler
Currently the resampler loops over each datapoint and variable j and resamples by:
  • Resamples a new point from a normal distribution centered at the original datapoint, with std=bw[j], for continuous varibales.
  • Resamples randomly from all catagories != current catagory if bw[j] > np.random.rand() for unordered catagorical variables.
  • Resamples using a Wang-Ryzin kernel defined at the datapoint using bandwith bw[j] for ordered catagorical variables.
ResampleEvaluation

Evaluates the distribution similarities based on Chi-square, Kolmogorov-Smirnov and the Cosine distance. The formulas and applications of these metrics to the datset are explained below.

Chi-square

When applying the two sample chi-square test we are testing whether two datasets come from a common distribution.

  • \(H_0\): The two sets come from a common distribution
  • \(H_1\): \(H_0\) is false, the sets come from different distributions

The Chi-square test we use is calculated using the formula:

\[\chi^2 = \sum_{i=1}^{k}{\frac{(K_R * R - K_E * E)^2}{R}}\]

where R is the resampled dataset and E the expected values, or in our context, the original dataset.

\[ \begin{align}\begin{aligned}K_R = \sqrt{\frac{\sum(E_i)}{\sum(R_i)}}\\K_E = \sqrt{\frac{\sum(R_i)}{\sum(E_i)}}\end{aligned}\end{align} \]

In case the datasets are not of equal sample size, we can still use the Chi-square test using the scaling constants. If the sets of of equal sample size, the constants will go to 1, and we are left with the ‘traditional’ chi-square formula:

\[\chi^2 = \sum_{i=1}^{k}{\frac{(R - E)^2}{E}}\]
_images/chi-plot.png
Kolmogorov-Smirnov

The Kolmogorov–Smirnov test may also be used to test whether two underlying one-dimensional probability distributions differ. In this case, we apply the KS test to each variable and save the results.

Note

The KS test assumes the distribution is continuous. Although the test is run for all variables, we should keep this in mind when looking at the results for the catagorical variables.

If the K-S statistic is small or the p-value is high, then we cannot reject the null-hypothesis that the distributions of the two samples are the same. Aka: They are sufficiently similar to say they are from the same distrubution.

\[\begin{split}F_n(x) = \frac{1}{n}\sum_{i=1}^{n}I_{[-\inf,x]}(X_i) \\ D_n = \sup_{n}|F_n(x) - F(x)|\end{split}\]
Cosine-distance

We also tried to define a distance from the original dataset to the resampled one. We employ the cosine distance applied to each point and its resampled point, represented as a vector. The distance will be 0 when on top of each other, and the max distance is.

We define a vector as the combination of all variables for one datapoint (or row in your dataset).
  • All continuous values are represented as is
  • All ordered catagorical values are mapped to numerical values going from 0 to # of catagories available, where 0 corresponds to the lowest ranking catagory.
  • All unordered catagorical are ignored for now since we have not yet defined a sufficient distance measure for these.

Then, the cosine distance is calculated for each point and it’s corresponding resample.

\[\cos(\theta) = \frac{A\cdot B}{\|A\|\|B\|}\]
Mimic Report
The mimic report link will create a standard eskapade style pdf report. The report includes per variable:
  • A stacked histogram plot showing before and after the resampling
  • A stacked histogram plot of the data per variable in the copula space and a normal distribution
  • A correlation matrix of numeric values before and after resimulation

Each variable page also contains the chi-square values comparing before and afer the resampling (also see Chi-square). For each variable, there is a table containing several values. The values correspond the chisquare calculation done on a 1D histogram of the variable itself, and done on 2D histograms of two variables as listed in the table.

Example: On the page of variable d

  Chi2 p-value dof
d 1.22018 0.269325 3
e 1034.82 0 3
f 317.124 0 3
g 1118.11 0 3
a 7.92157 0.0476607 3
b 1.4137 0.84181 3
c 1.43721 0.696837 3

The value 1.22 corresponds to the calculation variable d before and after the resampling. The value of 1034.82 corresponds to the calculations done on a 2D histogram of variables d and e, before and after the resampling.

Finally, two other metrics, the Kolmogorov-Smirnov and the cosine distance, are also calculated and shown in the report. You can find these on the last page.

All available examples

To see the available Eskapade examples, do:

$ export TUTDIRC=`pip show Eskapade-Core | grep Location | awk '{ print $2"/escore/tutorials" }'`
$ export TUTDIR=`pip show Eskapade | grep Location | awk '{ print $2"/eskapade/tutorials" }'`
$ ls -l $TUTDIRC/ $TUTDIR/

Many Eskapade example macros exist in the tutorials directory. The numbering of the example macros follows the package structure:

  • esk100+: basic macros describing the chains, links, and datastore functionality of Eskapade.
  • esk200+: macros describing links to do basic processing of pandas dataframes.
  • esk300+: visualization macros for making histograms, plots and reports of datasets.
  • esk500+: macros for doing data quality assessment and cleaning.
  • esk700+: macros for doing data simulation.

The Eskapade macros are briefly described below. They explain the basic architecture of Eskapade, i.e. how the chains, links, datastore, and process manager interact.

Hopefully you now have enough knowledge to run and explore Eskapade by yourself. You are encouraged to run all examples to see what Eskapade can do for you!

Example esk101: Hello World!

Macro 101 runs the Hello World Link. It runs the Link twice using a repeat kwarg, showing how to use kwargs in Links.

$ eskapade_run $TUTDIRC/esk101_helloworld.py
Example esk102: Multiple chains

Macro 102 uses multiple chains to print different kinds of output from one Link. This link is initialized multiple times with different kwargs and names. There are if-statements in the macro to control the usage of the chains.

$ eskapade_run $TUTDIRC/esk102_multiple_chains.py
Example esk103: Print the DataStore

Macro 103 has some objects in the DataStore. The contents of the DataStore are printed in the standard output.

$ eskapade_run $TUTDIRC/esk103_printdatastore.py
Example esk104: Basic DataStore operations

Macro 104 adds some objects from a dictionary to the DataStore and then moves or deletes some of the items. Next it adds more items and prints some of the objects.

$ eskapade_run $TUTDIRC/esk104_basic_datastore_operations.py
Example esk105: DataStore Pickling

Macro 105 has 3 versions: A, B and C. These are built on top of the basic macro esk105. Each of these 3 macro’s does something slightly different:

  • A does not store any output pickles,
  • B stores all output pickles,
  • C starts at the 3rd chain of the macro.

Using these examples one can see how the way macro’s are run can be controlled and what it saves to disk.

$ eskapade_run $TUTDIRC/esk105_A_dont_store_results.py
$ eskapade_run $TUTDIRC/esk105_B_store_each_chain.py
$ eskapade_run $TUTDIRC/esk105_C_begin_at_chain3.py
Example esk106: Command line arguments

Macro 106 shows us how command line arguments can be used to control the chains in a macro. By adding the arguments from the message inside of the macro we can see that the chains are not run.

$ eskapade_run $TUTDIRC/esk106_cmdline_options.py
Example esk107: Chain loop

Example 107 adds a chain to the macro and using a repeater Link it repeats the chain 10 times in a row.

$ eskapade_run $TUTDIRC/esk107_chain_looper.py
Example esk108: Event loop

Example 108 processes a textual data set, to loop through every word and do a Map and Reduce operation on the data set. Finally a line printer prints out the result.

$ source $TUTDIRC/esk108_eventlooper.sh
Example esk109: Debugging tips

This macro illustrates basic debugging features of Eskapade. The macro shows how to start a python session while running through the chains, and also how to break out of a chain.

$ eskapade_run $TUTDIRC/esk109_debugging_tips.py
Example esk110: Code profiling

This macro demonstrates how to run Eskapade with code profiling turned on.

$ eskapade_run $TUTDIRC/esk110_code_profiling.py
Example esk111: Loading a datastore from file

Macro illustrates how to load an external datastore from file.

$ eskapade_run $TUTDIRC/esk111_load_datastore_from_file.py
Example esk201: Read data

Macro 201 illustrates how to open files as pandas datasets. It reads a file into the DataStore. The first chain reads one csv into the DataStore, the second chain reads multiple files (actually the same file multiple times) into the DataStore. (Looping over data is shown in example esk209.)

$ eskapade_run $TUTDIR/esk201_readdata.py
Example esk202: Write data

Macro 202 illustrate writing pandas dataframes to file. It reads a DataFrame into the data store and then writes the DataFrame to csv format on the disk.

$ eskapade_run $TUTDIR/esk202_writedata.py
Example esk203: apply func to pandas df

Illustrates the link that calls basic apply() to columns of a pandas dataframes. See for more information pandas documentation:

http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.apply.html

$ eskapade_run $TUTDIR/esk203_apply_func_to_pandas_df.py
Example esk204: apply query to pandas df

Illustrates the link that applies basic queries to pandas dataframe. See for more information pandas documentation:

http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.query.html

$ eskapade_run $TUTDIR/esk204_apply_query_to_pandas_df.py
Example esk205: concatenate pandas dfs

Illustrates the link that calls basic concat() of pandas dataframes. See for more information pandas documentation:

http://pandas.pydata.org/pandas-docs/stable/merging.html

$ eskapade_run $TUTDIR/esk205_concatenate_pandas_dfs.py
Example esk206: merge pandas dfs

Illustrate link that calls basic merge() of pandas dataframes. For more information see pandas documentation:

http://pandas.pydata.org/pandas-docs/stable/merging.html

$ eskapade_run $TUTDIR/esk206_merge_pandas_dfs.py
Example esk207: record vectorizer

This macro performs the vectorization of an input column of an input dataframe. E.g. a columnn x with values 1, 2 is tranformed into columns x_1 and x_2, with values True or False assigned per record.

$ eskapade_run $TUTDIR/esk207_record_vectorizer.py
Example esk208: record factorizer

This macro performs the factorization of an input column of an input dataframe. E.g. a columnn x with values ‘apple’, ‘tree’, ‘pear’, ‘apple’, ‘pear’ is tranformed into columns x with values 0, 1, 2, 0, 2.

$ eskapade_run $TUTDIR/esk208_record_factorizer.py
Example esk209: read big data itr

Macro to that illustrates how to loop over multiple (possibly large!) datasets in chunks.

$ eskapade_run $TUTDIR/esk209_read_big_data_itr.py
Example esk210: dataframe restoration

Macro to illustrate writing pandas dataframes to file and reading them back in whilst retaining the datatypes and index using numpy and feather file formats.

$ eskapade_run $TUTDIR/esk210_dataframe_restoration.py
Example esk301: dfsummary plotter

Macro shows how to plot the content of a dataframe in a nice summary pdf file. (Example similar to $TUTDIR/tutorial_1.py.)

$ eskapade_run $TUTDIR/esk301_dfsummary_plotter.py
Example esk302: histogram_filler_plotter

Macro that illustrates how to loop over multiple (possibly large!) datasets in chunks, in each loop fill a (common) histogram, and plot the final histogram.

$ eskapade_run $TUTDIR/esk302_histogram_filler_plotter.py
Example esk303: histogrammar filler plotter

Macro that illustrates how to loop over multiple (possibly large!) datasets in chunks, in each loop fill a histogrammar histograms, and plot the final histograms.

$ eskapade_run $TUTDIR/esk302_histogram_filler_plotter.py
Example esk304: df boxplot

Macro shows how to boxplot the content of a dataframe in a nice summary pdf file.

$ eskapade_run $TUTDIR/esk304_df_boxplot.py
Example esk305: correlation summary

Macro to demonstrate generating nice correlation heatmaps using various types of correlation coefficients.

$ eskapade_run $TUTDIR/esk305_correlation_summary.py
Example esk306: concatenate reports

This macro illustrates how to concatenate the reports of several visualization links into one big report.

$ eskapade_run $TUTDIR/esk306_concatenate_reports.py
Example esk501: fix pandas dataframe

Macro illustrates how to call FixPandasDataFrame link that gives columns consistent names and datatypes. Default settings perform the following clean-up steps on an input dataframe:

  • Fix all column names. Eg. remove punctuation and strange characters, and convert spaces to underscores.
  • Check for various possible nans in the dataset, then make all nans consistent by turning them into numpy.nan (= float)
  • Per column, assess dynamically the most consistent datatype (ignoring all nans in that column). Eg. bool, int, float, datetime64, string.
  • Per column, make the data types of all rows consistent, by using the identified (or imposed) data type (by default ignoring all nans)
$ eskapade_run $TUTDIR/esk501_fix_pandas_dataframe.py
Example esk701: Mimic dataset

Macro that illustrates how to resample a dataset using kernel density estimation (KDE). The macro can handle contiunous data, and both ordered and unordered catagorical data. The macro is build up in the following way:

  • A dataset is simulated containing mixed data types, representing general input data.
  • Some cleaning steps are performed on the dataset
  • KDE is applied to the dataset
  • Using the estimated bandwidths of the KDE, the data is resampled
  • An evaluation is done on the resulting resimulated dataset
$ eskapade_run $TUTDIR/esk701_mimic_data.py
Example esk702: Mimic data only unordered

This macro illustrates how to resample an existing data set, containing only unordered catagorical data, using kernel density estimation (KDE) and a direct resampling technique.

$ eskapade_run $TUTDIR/esk702_mimic_data_only_unordered.py

Tips on coding

This section contains a general description on how to use Eskapade in combination with other tools, in particular for the purpose of developing code.

Eskapade in PyCharm

PyCharm is a very handy IDE for debugging Python source code. It can be used to run Eskapade stand-alone (i.e. like from the command line) and with an API.

Stand-alone
  • Install PyCharm on your machine.
  • Open project and point to the Eskapade source code
  • Configuration, in ‘Preferences’, check the following desired values:
    • Under ‘Project: eskapade’ / ‘Project Interpreter’:
      • The correct Python version (Python 3)
    • Under ‘Build, Execution & Deployment’ / ‘Console’ / ‘Python Console’:
      • The correct Python version (Python 3)
  • Install Eskapade in editable mode
  • Run/Debug Configuration:
    • Under ‘Python’ add new configuration
    • Script: path to the console script eskapade_run (located in the same directory as the interpreter specified above in ‘Project Interpreter’)
    • Script parameters: path to a macro to be debugged, e.g. $ESKAPADE/python/eskapade/tutorials/tutorial_1.py, and eskapade_run command line arguments, e.g. --begin-with=Summary
    • Python interpreter: check if it is the correct Python version (Python 3)

You should now be able to press the ‘play button’ to run Eskapade with the specified parameters.

Command Line Arguments

Overview

We start this section with a short overview of a few often used arguments of the Eskapade command eskapade_run. The only required argument is a configuration file, which can be a Python script (Eskapade macro) or a pickled Eskapade configuration object. This section gives an overview of the optional arguments of the run command.

At the end of running the Eskapade program, by default the DataStore and configuration object are pickled and written out to:

$ ls -l results/Tutorial_1/proc_service_data/v0/latest/

When you are working on a macro, once you are done tweaking it, you can also store the results of each chain in pickle files:

$ eskapade_run --store-all python/eskapade/tutorials/tutorial_1.py

Eskapade uses these pickle files to load the trained models and uses them to predict new samples real-time, but also to pick up running at a later stage in the chain setup.

For example, if running Eskapade takes a long time, you can run one chain as well:

$ eskapade_run --single-chain=Data python/eskapade/tutorials/tutorial_1.py

This command uses as input the stored pickle files from the previous chain. This might come in handy when, for example, data pre-processing of your training set takes a long time. In that case, you can run the pre-processing chain over night, store the results in a pickle file and start with the training chain the next day.

Start running Eskapade from a specified chain:

$ eskapade_run --begin-with=Summary python/eskapade/tutorials/tutorial_1.py

Stop running after a specified chain:

$ eskapade_run --end-with=Data python/eskapade/tutorials/tutorial_1.py

Below the most important command-line options are explained in detail.

Table of all arguments

The following table summarizes the available command-line options. Most of these options set variables in the Eskapade configuration object and can be overwritten by settings in the configuration macro.

Option Short option Argument Description
–help -h   show help message and exit
–analysis-name -n NAME set name of analysis in run
–analysis-version -v VERSION set version of analysis version in run
–batch-mode     run in batch mode (no X Windows)
–interactive -i   start Python shell after run
–log-level -L LEVEL set logging level
–log-format   FORMAT set log-message format
–unpickle-config     interpret first CONFIG_FILE as path to pickled settings
–profile     run profiler for Python code
–conf-var -c KEY=VALUE set configuration variable
–begin-with -b CHAIN_NAME begin execution with chain CHAIN_NAME
–end-with -e CHAIN_NAME end execution with chain CHAIN_NAME
–single-chain -s CHAIN_NAME only execute chain CHAIN_NAME
–store-all     store run-process services after every chain
–store-one   CHAIN_NAME store run-process services after chain CHAIN_NAME
–store-none     do not store run-process services
–results-dir   RESULTS_DIR set directory path for results output
–data-dir   DATA_DIR set directory path for data
–macros-dir   MACROS_DIR set directory path for macros
–templates-dir   TEMPLATES_DIR set directory path for template files
–spark-cfg-file   SPARK_CONFIG_FILE set path of Spark configuration file
–seed   SEED set seed for random-number generation

Description and examples

This section contains the most used options with a longer description of what it does and how it works combined with examples.

Set log level

The log level is controlled with the --log-level option. For example, to set the log level to “debug”, add:

--log-level=DEBUG

to the command line:

$ eskapade_run -L DEBUG python/eskapade/tutorials/tutorial_1.py

The available log levels are:

NOTSET,
DEBUG,
INFO,
WARNING,
ERROR,
FATAL

They correspond to the appropriate POSIX levels.

When writing your own Link, these levels can be accessed with the logger module:

self.logger.debug('Text to be printed when logging at DEBUG level')

All output is done in this manner, never with the python print function, since this yields us more control over the process.

Help

Help can be called by running the following:

$ eskapade_run --help
Interactive python mode

To keep the results in memory at end of session and access them in an interactive session, run Eskapade in interactive mode. This is controlled with --interactive:

$ eskapade_run -i python/eskapade/tutorials/tutorial_1.py

At the end of the session a Python console is started from which e.g. the data store can be accessed.

Saving states

To write out the intermediate results from every chain, add the command line argument --store-all. This will write pickles in results/NAME/proc_service_data/VERSION/, containing the state of Eskapade at the end of the chain:

$ eskapade_run --store-all python/eskapade/tutorials/tutorial_1.py

To write out the state after a particular chain, use option --store-one:

$ eskapade_run --store-one=Data python/eskapade/tutorials/tutorial_1.py

To not store any pickle files, run with the option --store-none:

$ eskapade_run --store-none python/eskapade/tutorials/tutorial_1.py
Single Chain

To run a single chain, use the option --single-chain. This picks up the data stored by the previous chain in the macro. It is, therefore, necessary to have run the previous chain, otherwise the engine can not start:

$ eskapade_run -s Summary python/eskapade/tutorials/tutorial_1.py
Start from a Chain

To start from a chain use the command line argument --begin-with. This picks up the data stored by the previous chain in the macro.

$ eskapade_run -b Summary python/eskapade/tutorials/tutorial_1.py
Stop at a Chain

To end the running of the engine at a chain use, the command line argument --end-with:

$ eskapade_run -e Data python/eskapade/tutorials/tutorial_1.py
Changing analysis version

A version number is assigned to each analysis, which by default is 0. It can be upgraded by using the option --analysis-version. When working on an analysis, it is recommended to update this number regularly for bookkeeping purposes. The command line always has higher priority over the macro. If the macro is version 0 and the command line uses version 1, the command line will overrule the macro.

$ eskapade_run -v 1 python/eskapade/tutorials/tutorial_1.py

Notice that the output of this analysis is now stored in the directory:

$ ls -l results/Tutorial_1/data/v1/report/

Notice as well that, for bookkeeping purposes, a copy of the (evolving) configuration macro is always stored as well, under:

$ ls -l results/Tutorial_1/config/v1/tutorial_1.py
Running an old configuration (macro)

Settings for the Eskapade run are stored in a configuration object, which is accessed as a run-process service. This run-time service can be persisted as a file, which is normally done at the end of the run.

Persisted settings can be used in a following run by providing the file path of the ConfigObject pickle file as the configuration file argument. The option --unpickle-config is required to indicate that this file contains persisted settings:

$ eskapade_run --unpickle-config results/Tutorial_1/proc_service_data/v0/latest/eskapade.core.process_services.ConfigObject.pkl

In this way, rolling back to a previous point is straight-forward.

For lookup purposes a copy of the configuration macro is always stored as well, under:

$ ls -l results/Tutorial_1/config/v0/tutorial_1.py
Profiling your code

Your can profile the execution of your analysis functions with the option --profile:

$ eskapade_run --profile=cumulative python/eskapade/tutorials/tutorial_1.py

After running this prints out a long list of all functions called, including the time it took to run each of of them, where the functions are sorted based on cumulative time.

To get the the list of sorting options for the profiling, run:

$ eskapade_run --help
Combining arguments

Of course you can add multiple arguments to the command line, the result would be for example an interactive session in debug mode that writes out intermediate results from each chain:

$ eskapade_run -i --store-all -L DEBUG -c do_chain0=False -c mydict="{'f': 'y=pi', 'pi': 3.14}" python/eskapade/tutorials/esk106_cmdline_options.py

Package structure

Eskapade contains many tools, and to find and use them most efficiently it is necessary to understand how the repository is build up. This section discusses the structure of the code and how the framework handles subpackages.

Architecture

The architecture of Eskapade can be summarized in this picture:

_images/eskapade_architecture.png

The example we just discussed generally shows how the framework works. The steps it takes are the following:

  • eskapade_run configures Eskapade based on user settings and/or the macro file, and executes all chains and links;
  • Macros (python file) contain Chains;
  • Chains (python object) contains Links;
  • Links (python class) contain analysis code.

The chains are run in the order of ‘registering’ them in the ProcessManager.

The ProcessManager is the ultimate object that executes all the code in your macro. It also keeps track of the configuration of Eskapade, and of the objects in the data store that are passable between links.

The settings flow of Eskapade is shown in the following picture:

_images/Eskapade_8_4_settings_flow.png

The components of the architecture of Eskapade are explained in further detail in the Tutorials section.

Structure

When using Eskapade it is important to understand where all components are located. The components can be for example links or utilities that you want to use.

The Eskapade framework is contained in the Python package eskapade, which lives in the python directory. Every specific subject has its subpackage in eskapade, containing the utilities it needs, the links that are defined for the subject.

The core of the framework is implemented in the core subpackage. This subpackage contains the low-level machinery for running analysis algorithms in chains of links. The core_ops subpackage contains basic links for operating this framework.

An example of a typical subpackage is eskapade.analysis, which contains basic analysis tools. Its structure is common to all Eskapade subpackages:

|-eskapade
   |-analysis
      |-links

The subpackage contains several modules, which contain classes and functions to be applied in links. The eskapade.analysis.statistics module, for example, contains code to generate an overview of the statistical properties of variables in given input data.

Eskapade links are located in the links directory. There is a separate module for each link, defining the link class instance. By convention, the names of the module and class are both the link name, the former in snake case and the latter in camel case. For example, the module read_to_df defines the link class ReadToDf.

The tests are contained separately in the Python package eskapade_python under tests directory. Ideally, there is a test module for each (link) module in the Eskapade package. Optionally, integration tests are implemented in integration. For the eskapade.analysis package, there is the module test_tutorial_macros with integration tests that run the tutorial macros corresponding to this subpackage:

|-eskapade_python
   |-analysis
      |-integration
         |-test_tutorial_macros.py

Subpackages

Eskapade contains the following list of subpackages:

  • core is the package that contains the core framework of Eskapade.
  • core_ops contains links pertaining to the core functionality of Eskapade.
  • analysis contains pandas links and code.
  • visualization contains visualization code and plotter links.
  • data_quality contains links and code for fixing messy data.
  • data_mimic contains links and code for re-simulating data with mixed data-types

Imports

Main elements of the Eskapade framework are imported directly from the eskapade package. For example, the run-configuration object and the run-process manager are part of the core subpackage, but are imported by

from eskapade import process_manager, ConfigObject

Links are imported directly from their subpackage:

from eskapade.analysis import ReadToDf

In a macro, you can now instantiate and configure the ReadToDf link and add it to a chain in the process manager.

Results

Results of a macro are written out by default in the results directory. The analysis run is persisted in the results directory by the analysis_name given in the macro. This directory has the following structure:

  • config: the configuration macro
  • proc_service_data: persisted states of run-process services
  • data: analysis results, such as graphs or a trained model

The data for each of these elements are stored by the analysis version, e.g. v0, v1, v2, etc. For example, the report produced by the tutorial esk301_dfsummary_plotter is saved in the directory results/esk301_dfsummary_plotter/data/v0/report.

Debugging

When building new Links or other functionality you will want to debug at some point. There are multiple ways to do this, because there are multiple ways of running the framework. A few ways are:

  • Running in the terminal. In this scenario you have to work in a virtual environment (or adjust your own until it has all dependencies) and debug using the terminal output.
  • Running in a notebook. This way the code is run in a notebook and you can gather the output from the browser.
  • Running in a docker. The code is run in the docker and the repository is mounted into the container. The docker (terminal) returns output.
  • Running in a VM. In this case you run the code in the VM and mount the code into the VM. The output can be gathered in the VM and processed in the VM.

In the first three options you want to use an IDE or text-editor in a ‘normal’ environment to debug your code and in the last option you can use an editor in the VM or outside of it.

Troubleshooting

The least error prone ways are docker and VMs, because they automatically have the dependencies set.

Release notes

Version 0.9.0

Eskapade release v0.9.0 has the following useful upgrades:

  • The core functionality of Eskapade, namely: the Link, Chain, process_manager, DataStore, ConfigObject and corresponding tutorials, have been split off from the growing (ever more analysis related) Eskapade repository, into the new package Eskapade-Core. Eskapade-Core is a very light-weight Python3 package.

  • A new module data_mimic has been add to Eskapade, including tutorials, meant for resimulating existing datasets.

  • We have added feather i/o functionality for reading and writeng dataframes.

  • The logger has been fixed, it is now possible to set the log-level of loggers again.

  • The Eskapade docker files have been taken out of the Eskapade repository to avoid version conflicts, into the new git repo Eskapade-Environment.

  • The Eskapade docker image eskapade-usr contain the latest working versions of Eskapade, Eskapade-Core, Eskapade-ROOT, and Eskapade-Spark. Type:

    $ docker pull kave/eskapade-usr:latest

    to pull it in.

Version 0.8.2

The Eskapade patch release v0.8.2 and corresponding docker containers fix two issues:

  • The matplotlib backend is no longer set to batchmode when running Eskapade in a jupyter notebook. By default, batch mode is only turned on when no DISPLAY environment variable is set, and when not running in a notebook; the batch-mode flag can also be controlled with the command line option –batch-mode.

  • The Eskapade docker containers contain working version of Eskapade, Eskapade-ROOT, and Eskapade-Spark. Type:

    $ docker pull kave/eskapade-usr:0.8.2

    to pull it in.

Version 0.8

In version 0.8 of Eskapade (August 2018) the modules root-analysis and spark-analysis have been split off into separate packages called Eskapade-ROOT and Eskapade-Spark .

So the (core) Eskapade package no longer depends on ROOT and Spark, just on plain python packages. This make it much easier for people to try out the core functionality of Eskapade.

  • To install Eskapade-ROOT and Eskapade-Spark, do:

    $ pip install Eskapade-ROOT
$ pip install Eskapade-Spark

    or check out the code from out github repository:

    $ git clone https://github.com/KaveIO/Eskapade-ROOT.git eskapade-root
$ pip install -e eskapade-root/
$ git clone https://github.com/KaveIO/Eskapade-Spark.git eskapade-spark
$ pip install -e eskapade-spark/

    where in this example the code is installed in edit mode (option -e).

    You can now use these in Python with:

    import eskapadespark
import esroofit

Version 0.7

Version 0.7 of Eskapade (February 2018) contains several major updates:

  • The Eskapade code has been made pip friendly. One can now simply do:

    $ pip install Eskapade

    or check out the code from out github repository:

    $ git clone https://github.com/KaveIO/Eskapade.git
$ pip install -e Eskapade/

    where in this example the code is installed in edit mode (option -e).

    You can now use Eskapade in Python with:

    import eskapade

    This change has resulted in some restructuring of the python directories, making the overall structure more transparent: all python code, including the tutorials, now fall under the (single) python/ directory. Additionally, thanks to the pip convention, our prior dependence on environment variables ($ESKAPADE) has now been fully stripped out of the code.

  • There has been a cleanup of the core code, removing obsolete code and making it better maintainable. This has resulted in a (small) change in the api of the process manager, adding chains, and using the logger. All tutorials and example macro files have been updated accordingly. See the migration section for detailed tips on migrating existing Eskapade code to version 0.7.

  • All eskapade commands now start with the prefix eskapade_. All tutorials have been updated accordingly. We have the commands:

    • eskapade_bootstrap, for creating a new Eskapade analysis project. See this new tutorial for all the details.
    • eskapade_run, for running the Eskapade macros.
    • eskapade_trail, for running the Eskapade unit and integration tests.
    • eskapade_generate_link, eskapade_generate_macro, eskapade_generate_notebook, for generating a new link, macro, or Jupyter notebook respectively.

Version 0.6

The primary feature of version 0.6 (August 2017) is the inclusion of Spark, but this version also includes several other new features and analyses.

We include multiple Spark links and 10 Spark examples on:

  • The configuration of Spark, reading, writing and converting Spark dataframes, applying functions and queries to dataframes, filling histograms and (very useful!) applying arbitrary functions (e.g. pandas) to groupby calls.

In addition we hade added:

  • A ROOT analysis for studying and quantifying between sets of (non-)categorical and observables. This is useful for finding outliers in arbitrary datasets (e.g. surveys), and we include a tutorial of how to do this.
  • A ROOT analysis on predictive maintenance that decomposes a distribution of time difference between malfunctions by fitting this multiple Weibull distributions.
  • New flexible features to create and chain analysis reports with several analysis and visualization links.

Version 0.5

Our 0.5 release (May 2017) contains multiple new features, in particular:

  • Support for ROOT, including multiple examples on using data analysis, fitting and simulation examples using RooFit.
  • Histogram conversion and filling support, using ROOT, numpy, Histogrammar and Eskapade-internal histograms.
  • Automated data-quality fixes for buggy columns datasets, including data type fixing and NaN conversion.
  • New visualization utilities, e.g. plotting multiple types of (non-linear) correlation matrices and dendograms.
  • And most importantly, many new and interesting example macros illustrating the new features above!

Version 0.4

In our 0.4 release (Feb 2017) we are releasing the core code to run the framework. It is written in python 3. Anyone can use this to learn Eskapade, build data analyses with the link-chain methodology, and start experiencing its advantages.

The focus of the provided documentation is on constructing a data analysis setup in Eskapade. Machine learning interfaces will be included in an upcoming release.

Developing and Contributing

Preliminaries

Working on Eskapade

You have some cool feature and/or algorithm you want to add to Eskapade. How do you go about it?

First clone Eskapade.

git clone https://github.com/KaveIO/Eskapade.git eskapade

then

pip install -e eskapade

this will install Eskapade in editable mode, which will allow you to edit the code and run it as you would with a normal installation of eskapade.

To make sure that everything works try executing eskapade without any arguments, e.g.

eskapade_run

or you could just execute the tests using either the eskapade test runner, e.g.

cd eskapade
eskapade_trial .

or

cd eskapade
python setup.py test

That’s it.

Contributing

When contributing to this repository, please first discuss the change you wish to make via issue, email, or any other method with the owners of this repository before making a change. You can find the contact information on the index page.

Note that when contributing that all tests should succeed.

Tips and Tricks

  • Enable auto reload in ipython:
%load_ext autoreload

this will reload modules before executing any user code.

API

API Documentation

Eskapade
eskapade package
Subpackages
eskapade.analysis package
Subpackages
Submodules
eskapade.analysis.correlation module

Project: Eskapade - A python-based package for data analysis.

Created: 2018/06/23

Description:

Correlation related util functions.

Convert Pearson correlation value into a chi2 value of a contingency test matrix of a bivariate gaussion, and vice-versa. Calculation uses scipy’s mvn library. Calculates correlation coëfficients based on mutual_information, correlation_ratio, pearson, kendall or spearman methods.

Authors:
KPMG Advanced Analytics & Big Data team, Amstelveen, The Netherlands

Redistribution and use in source and binary forms, with or without modification, are permitted according to the terms listed in the file LICENSE.

eskapade.analysis.correlation.calculate_correlations(df, method)

Calculates correlation coefficients between every column pair.

Parameters:
  • df (pd.DataFrame) – input data frame
  • method (str) – mutual_information, correlation_ratio, pearson, kendall or spearman, phik, significance
Returns:

pd.DataFrame
eskapade.analysis.correlation.chi2_from_rho(rho, n, subtract_from_chi2=0, corr0=None, sx=None, sy=None, nx=-1, ny=-1)

Calculate chi2-value of bivariate gauss having correlation value rho

Calculate no-noise chi2 value of bivar gauss with correlation rho, with respect to bivariate gauss without any correlation.

Returns float:chi2 value
eskapade.analysis.correlation.rho_from_chi2(chi2, n, nx, ny, sx=None, sy=None)

correlation coefficient of bivariate gaussian derived from chi2-value

Chi2-value gets converted into correlation coefficient of bivariate gauss with correlation value rho, assuming giving binning and number of records. Correlation coefficient value is between 0 and 1.

Bivariate gaussian’s range is set to [-5,5] by construction.

Returns float:correlation coefficient
eskapade.analysis.datetime module

Project: Eskapade - A python-based package for data analysis.

Classes: TimePeriod, FreqTimePeriod

Created: 2017/03/14

Description:
Time period and time period with frequency.
Authors:
KPMG Advanced Analytics & Big Data team, Amstelveen, The Netherlands

modification, are permitted according to the terms listed in the file Redistribution and use in source and binary forms, with or without LICENSE.

class eskapade.analysis.datetime.FreqTimePeriod(**kwargs)

Bases: eskapade.analysis.datetime.TimePeriod

Time period with frequency.

__init__(**kwargs)

Initialize TimePeriod instance.

dt_string(period_index)

Convert period index into date/time string (start of period).

Parameters:period_index (int) – specified period index value.
freq

Return frequency.

period_index(dt)

Return number of periods until date/time “dt” since 1970-01-01.

Parameters:dt – specified date/time parameter
class eskapade.analysis.datetime.TimePeriod(**kwargs)

Bases: escore.core.mixin.ArgumentsMixin

Time period.

__init__(**kwargs)

Initialize TimePeriod instance.

logger

A logger that emits log messages to an observer.

The logger can be instantiated as a module or class attribute, e.g.

>>> logger = Logger()
>>> logger.info("I'm a module logger attribute.")
>>>
>>> class Point(object):
>>>     logger = Logger()
>>>
>>>     def __init__(self, x = 0.0, y = 0.0):
>>>         Point.logger.debug('Initializing {point} with x = {x}  y = {y}', point=Point, x=x, y=y)
>>>         self._x = x
>>>         self._y = y
>>>
>>>     @property
>>>     def x(self):
>>>         self.logger.debug('Getting property x = {point._x}', point=self)
>>>         return self._x
>>>
>>>     @x.setter
>>>     def x(self, x):
>>>         self.logger.debug('Setting property y = {point._x}', point=self)
>>>         self._x = x
>>>
>>>     @property
>>>     def y(self):
>>>        self.logger.debug('Getting property y = {point._y}', point=self)
>>>        return self._y
>>>
>>>     @y.setter
>>>     def y(self, y):
>>>         self.logger.debug('Setting property y = {point._y}', point=self)
>>>         self._y = y
>>>
>>> a_point = Point(1, 2)
>>>
>>> logger.info('p_x = {point.x} p_y = {point.y}', point=a_point)
>>> logger.log_level = LogLevel.DEBUG
>>> logger.info('p_x = {point.x} p_y = {point.y}', point=a_point)

The logger uses PEP-3101 (Advanced String Formatting) with named placeholders, see <https://www.python.org/dev/peps/pep-3101/> and <https://pyformat.info/> for more details and examples.

Furthermore, logging events are only formatted and evaluated for logging levels that are enabled. So, there’s no need to check the logging level before logging. It’s also efficient.

classmethod parse_date_time(dt)

Try to parse specified date/time.

Parameters:dt – specified date/time
classmethod parse_time_period(period)

Try to parse specified time period.

Parameters:period – specified period
period_index(dt)

Get number of periods until date/time “dt”.

Parameters:dt – specified date/time
class eskapade.analysis.datetime.UniformTsTimePeriod(**kwargs)

Bases: eskapade.analysis.datetime.TimePeriod

Time period with offset.

__init__(**kwargs)

Initialize TimePeriod instance.

offset

Get offset parameter.

period

Get period parameter.

period_index(dt)

Get number of periods until date/time “dt” since “offset”, given specified “period”.

Parameters:dt – specified date/time
eskapade.analysis.histogram module

Project: Eskapade - A python-based package for data analysis.

Classes: ValueCounts, BinningUtil, Histogram

Created: 2017/03/14

Description:
Generic 1D Histogram class.
Authors:
KPMG Advanced Analytics & Big Data team, Amstelveen, The Netherlands

modification, are permitted according to the terms listed in the file Redistribution and use in source and binary forms, with or without LICENSE.

class eskapade.analysis.histogram.BinningUtil(**kwargs)

Bases: object

Helper for interpreting bin specifications.

BinningUtil is a helper class used for interpreting bin specification dictionaries. It is a base class for the Histogram class.

__init__(**kwargs)

Initialize link instance.

A bin_specs dictionary needs to be provided as input. bins_specs is a dict containing ‘bin_width’ and ‘bin_offset’ keys. In case bins widths are not equal, bin_specs contains ‘bin_edges’ (array) instead of ‘bin_width’ and ‘bin_offset’. ‘bin_width’ and ‘bin_offset’ can be numeric or numpy timestamps.

Alternatively, bin_edges can be provided as input to bin_specs.

Example bin_specs dictionaries are:

>>> bin_specs = {'bin_width': 1, 'bin_offset': 0}
>>> bin_spect = {'bin_edges': [0, 2, 3, 4, 5, 7, 8]}
>>> bin_specs = {'bin_width': np.timedelta64(30, 'D'),
                 'bin_offset': np.datetime64('2010-01-04')}
Parameters:
  • bin_specs (dict) – dictionary contains ‘bin_width’ and ‘bin_offset’ numbers or ‘bin_edges’ array. Default is None.
  • bin_edges (list) – array with numpy histogram style bin_edges. Default is None.
bin_specs

Get bin_specs dictionary.

Returns:bin_specs dictionary
Return type:dict
get_bin_center(bin_label)

Return bin center for a given bin index.

Parameters:bin_label – bin label for which to find the bin center
Returns:bin center, can be float, int, timestamp
get_bin_edges()

Return bin edges.

Returns:bin edges
Return type:array
get_bin_edges_range()

Return bin range determined from bin edges.

Returns:bin range
Return type:tuple
get_left_bin_edge(bin_label)

Return left bin edge for a given bin index.

Parameters:bin_label – bin label for which to find the left bin edge
Returns:bin edge, can be float, int, timestamp
get_right_bin_edge(bin_label)

Return right bin edge for a given bin index.

Parameters:bin_label – bin label for which to find the right bin edge.
Returns:bin edge, can be float, int, timestamp
truncated_bin_edges(variable_range=None)

Bin edges corresponding to a given variable range.

Parameters:variable_range (list) – variable range used for finding the right bin edges array. Optional.
Returns:truncated bin edges
Return type:array
value_to_bin_label(var_value, greater_equal=False)

Return bin index for given bin value.

Parameters:
  • var_value – variable value for which to find the bin index
  • greater_equal (bool) – for float, int, timestamp, return index of bin for which value falls in range [lower edge, upper edge). If set to true, return index of bin for which value falls in range [lower edge, upper edge]. Default if false.
Returns:

bin index
Return type:

int
class eskapade.analysis.histogram.Histogram(counts, **kwargs)

Bases: eskapade.analysis.histogram.BinningUtil, escore.core.mixin.ArgumentsMixin

Generic 1D Histogram class.

Histogram holds bin labels (name of each bin), value_counts (values of the histogram) and a variable name. The bins can be categoric or numeric, where numeric includes timestamps. In case of numeric bins, bin_specs is set. bins_specs is a dict containing bin_width and bin_offset. In case bins widths are not equal, bin_specs contains bin_edges instead of bin_width and bin_offset.

__init__(counts, **kwargs)

Initialize Histogram instance.

A bin_specs dictionary can be provided as input. bins_specs is a dict containing ‘bin_width’ and ‘bin_offset’ keys. In case bins widths are not equal, bin_specs contains ‘bin_edges’ (array) instead of ‘bin_width’ and ‘bin_offset’. ‘bin_width’ and ‘bin_offset’ can be numeric or numpy timestamps.

Histogram counts can be specified as a ValueCounts object, a dictionary or a tuple:

  • tuple: Histogram((bin_values, bin_edges), variable=<your_variable_name>)
  • dict: a dictionary as comes out of pandas.series.value_counts() or pandas.Dataframe.groupby.size() over one variable.
  • ValueCounts: a ValueCounts object contains a value_counts dictionary.

Example bin_specs dictionaries are:

>>> bin_specs = { 'bin_width': 1, 'bin_offset': 0 }
>>> bin_spect = { 'bin_edges': [0,2,3,4,5,7,8] }
>>> bin_specs = { 'bin_width': np.timedelta64(30,'D'),
                  'bin_offset': np.datetime64('2010-01-04') }
Parameters:
  • counts – histogram counts
  • bin_specs (dict) – dictionary contains ‘bin_width’ and ‘bin_offset’ numbers or ‘bin_edges’ array (default is None)
  • variable (str) – name of the variable represented by the histogram
  • datatype (type) – data type of the variable represented by the histogram (optional)
bin_centers()

Return bin centers.

Returns:array of the bin centers
Return type:array
bin_edges()

Return numpy style bin_edges array with uniform binning.

Returns:array of all bin edges
Return type:array
bin_entries()

Return number of bin entries.

Return the bin counts of the known bins in the value_counts object.

Returns:array of the bin counts
Return type:array
bin_labels()

Return bin labels.

Returns:array of all bin labels
Return type:array
classmethod combine_hists(hists, labels=False, rel_bin_width_tol=1e-06, **kwargs)

Combine a set of histograms.

Parameters:
  • hists (array) – array of Histograms to add up.
  • labels (label) – histograms to add up have labels? (else are numeric) Default is False.
  • variable (str) – name of variable described by the summed-up histogram
  • rel_bin_width_tol (float) – relative tolerance between numeric bin edges.
Returns:

summed up histogram
Return type:

Histogram
copy(**kwargs)

Return a copy of this histogram.

Parameters:variable (str) – assign new variable name
datatype

Data type of the variable represented by the histogram.

Returns:data type
Return type:type
get_bin_count(bin_label)

Get bin count for specific bin label.

Parameters:bin_label – a specific key to find corresponding bin.
Returns:bin counter value
Return type:int
get_bin_labels()

Return all bin labels.

Returns:array of all bin labels
Return type:array
get_bin_range()

Return the bin range.

Returns:tuple of the bin range found
Return type:tuple
get_bin_vals(variable_range=None, combine_values=True)

Get bin labels/edges and corresponding bin counts.

Bin values corresponding to a given variable range.

Parameters:
  • variable_range (list) – variable range used for finding the right bins to get values from. Optional.
  • combine_values (bool) – if bin_specs is not set, combine existing bin labels with variable range.
Returns:

two arrays of bin values and bin edges
Return type:

array
get_hist_val(var_value)

Get bin count for bin by value of histogram variable.

Parameters:var_value – a specific value to find corresponding bin.
Returns:bin counter value
Return type:int
get_nonone_bin_centers()

Return bin centers.

Returns:array of the bin centers
Return type:array
get_nonone_bin_counts()

Return bin counts.

Returns:array of the bin counts
Return type:array
get_nonone_bin_edges()

Return numpy style bin-edges array.

Returns:array of the bin edges
Return type:array
get_nonone_bin_range()

Return the bin range.

Returns:tuple of the bin range found
Return type:tuple
get_uniform_bin_edges()

Return numpy style bin-edges array with uniform binning.

Returns:array of all bin edges
Return type:array
logger

A logger that emits log messages to an observer.

The logger can be instantiated as a module or class attribute, e.g.

>>> logger = Logger()
>>> logger.info("I'm a module logger attribute.")
>>>
>>> class Point(object):
>>>     logger = Logger()
>>>
>>>     def __init__(self, x = 0.0, y = 0.0):
>>>         Point.logger.debug('Initializing {point} with x = {x}  y = {y}', point=Point, x=x, y=y)
>>>         self._x = x
>>>         self._y = y
>>>
>>>     @property
>>>     def x(self):
>>>         self.logger.debug('Getting property x = {point._x}', point=self)
>>>         return self._x
>>>
>>>     @x.setter
>>>     def x(self, x):
>>>         self.logger.debug('Setting property y = {point._x}', point=self)
>>>         self._x = x
>>>
>>>     @property
>>>     def y(self):
>>>        self.logger.debug('Getting property y = {point._y}', point=self)
>>>        return self._y
>>>
>>>     @y.setter
>>>     def y(self, y):
>>>         self.logger.debug('Setting property y = {point._y}', point=self)
>>>         self._y = y
>>>
>>> a_point = Point(1, 2)
>>>
>>> logger.info('p_x = {point.x} p_y = {point.y}', point=a_point)
>>> logger.log_level = LogLevel.DEBUG
>>> logger.info('p_x = {point.x} p_y = {point.y}', point=a_point)

The logger uses PEP-3101 (Advanced String Formatting) with named placeholders, see <https://www.python.org/dev/peps/pep-3101/> and <https://pyformat.info/> for more details and examples.

Furthermore, logging events are only formatted and evaluated for logging levels that are enabled. So, there’s no need to check the logging level before logging. It’s also efficient.

n_bins

Number of bins in the ValueCounts object.

Returns:number of bins
Return type:int
n_dim

Number of histogram dimensions.

The number of histogram dimensions, which is equal to one by construction.

Returns:number of dimensions
Return type:int
num_bins

Number of bins.

Returns:number of bins
Return type:int
remove_keys_of_inconsistent_type(prefered_key_type=None)

Remove all keys that have inconsistent data type(s).

Parameters:prefered_key_type (tuple) – the prefered key type to keep. Can be a tuple, list, or single type. E.g. str or (int,str,float). If None provided, the most common key type found is kept.
simulate(size, *args)

Simulate data using self (Histogram instance) as PDF.

see https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.rv_continuous.html

Parameters:size (int) – number of data points to generate
Return numpy.array generated_data:
 the generated data
Returns:Histogram of the generated data
Return type:Histogram
surface()

Calculate surface of the histogram.

Returns:surface
to_normalized(**kwargs)

Return a normalized copy of this histogram.

Parameters:
  • new_var_name (str) – assign new variable name
  • variable_range (list) – variable range used for finding the right bins to get values from.
  • combine_values (bool) – if bin_specs is not set, combine existing bin labels with variable range.
variable

Name of variable represented by the histogram.

Returns:variable name
Return type:string
class eskapade.analysis.histogram.ValueCounts(key, subkey=None, counts=None, sel=None)

Bases: object

A dictionary of value counts.

The dictionary of value counts comes out of pandas.series.value_counts() for one variable or pandas.Dataframe.groupby.size() performed over one or multiple variables.

__init__(key, subkey=None, counts=None, sel=None)

Initialize link instance.

Parameters:
  • key (list) – key is a tuple, list or string of (the) variable name(s), matching those and the structure of the keys in the value_counts dictionary.
  • subkey (list) – subset of key. If provided, the value_counts dictionary will be projected from key onto the (subset of) subkey. E.g. use this to map a two dimensional value_counts dictionary onto one specified dimension. Default is None. Optional.
  • counts (dict) – the value_counts dictionary.
  • sel (dict) – Apply selections to value_counts dictionary. Default is {}. Optional.
count(value_bin)

Get bin count for specific bin-key value bin.

Parameters:value_bin (tuple) – a specific key, and can be a list or tuple.
Returns:specific bin counter value
Return type:int
counts

Process and return value-counts dictionary.

Returns:after processing, returns the value_counts dictionary
Return type:dict
create_sub_counts(subkey, sel=None)

Project existing value counts onto a subset of keys.

E.g. map variables x,y onto single dimension x, so for each bin in x integrate over y.

Parameters:
  • subkey (tuple) – input sub-key, is a tuple, list, or string. This is the new key of variables for the returned ValueCounts object.
  • sel (dict) – dictionary with selection. Optional.
Returns:

value_counts object where subkey has become the new key.
Return type:

ValueCounts
get_values(val_keys=())

Get all key-values of a subset of keys.

E.g. give all x values in of the keys, when the value_counts object has keys (x, y).

Parameters:val_keys (tuple) – a specific sub-key to get key values for.
Returns:all key-values of a subset of keys.
Return type:tuple
key

Process and return current value-counts key.

Returns:the key
Return type:tuple
nononecounts

Return value-counts dictionary without None keys.

Returns:after processing, returns the value_counts dictionary without None keys
Return type:dict
num_bins

Number of value-counts bins.

Returns:number of bins
Return type:int
num_nonone_bins

Number of not-none value-counts bins.

Returns:number of not-none bins
Return type:int
process_counts(accept_equiv=True)

Project value counts onto the existing subset of keys.

E.g. map variables x,y onto single dimension x, so for each bin in x integrate over y.

Parameters:accept_equiv (bool) – accept equivalence of key and subkey if if subkey is in different order than key. Default is true.
Returns:successful projection or not
Return type:bool
remove_keys_of_inconsistent_type(prefered_key_type=None)

Remove keys with inconsistent data type(s).

Parameters:prefered_key_type (tuple) – the prefered key type to keep. Can be a tuple, list, or single type. E.g. str or (int, str, float). If None provided, the most common key type found is kept.
skey

Current value-counts subkey.

Returns:the subkey
Return type:tuple
sum_counts

Sum of counts of all value-counts bins.

Returns:the sum of counts of all bins
Return type:float
sum_nonone_counts

Sum of not-none counts of all value-counts bins.

Returns:the sum of not-none counts of all bins
Return type:float
eskapade.analysis.histogram_filling module

Project: Eskapade - A python-based package for data analysis.

Class: HistogramFillerBase

Created: 2017/03/21

Description:
Algorithm to fill histogrammar sparse-bin histograms. It is possible to do cleaning of these histograms by rejecting certain keys or removing inconsistent data types. Timestamp columns are converted to nanoseconds before the binning is applied.
Authors:
KPMG Advanced Analytics & Big Data team, Amstelveen, The Netherlands

Redistribution and use in source and binary forms, with or without modification, are permitted according to the terms listed in the file LICENSE.

class eskapade.analysis.histogram_filling.HistogramFillerBase(**kwargs)

Bases: escore.core.element.Link

Base class link to fill histograms.

It is possible to do after-filling cleaning of these histograms by rejecting certain keys or removing inconsistent data types. Timestamp columns are converted to nanoseconds before the binning is applied. Final histograms are stored in the datastore.

__init__(**kwargs)

Initialize link instance.

Store and do basic check on the attributes of link HistogramFillerBase.

Parameters:
  • name (str) – name of link
  • read_key (str) – key of input data to read from data store
  • store_key (str) – key of output data to store histograms in data store
  • columns (list) – colums to pick up from input data. (default is all columns)
  • bin_specs (dict) – dictionaries used for rebinning numeric or timestamp columns

Example bin_specs dictionary is:

>>> bin_specs = {'x': {'bin_width': 1, 'bin_offset': 0},
                 'y': {'bin_edges': [0, 2, 3, 4, 5, 7, 8]}}
Parameters:
  • var_dtype (dict) – dict of datatypes of the columns to study from dataframe. If not provided, try to determine datatypes directy from dataframe.
  • store_at_finalize (bool) – Store histograms in datastore at finalize(), not at execute(). Useful when looping over datasets. Default is False.
  • dict (drop_keys) – dictionary used for dropping specific keys from bins dictionaries of histograms

Example drop_keys dictionary is:

>>> drop_keys = {'x': [1,4,8,19],
                 'y': ['apple', 'pear', 'tomato'],
                 'x:y': [(1, 'apple'), (19, 'tomato')]}
assert_dataframe(df)

Check that input data is a filled pandas data frame.

Parameters:df – input (pandas) data frame
categorize_columns(df)

Categorize columns of dataframe by data type.

Parameters:df – input (pandas) data frame
drop_requested_keys(name, counts)

Drop requested keys from counts dictionary.

Parameters:
  • name (string) – key of drop_keys dict to get array of keys to be dropped
  • counts (dict) – counts dictionary to drop specific keys from
Returns:

count dict without dropped keys
execute()

Execute the link.

Execute() four things:

  • check presence and data type of requested columns
  • timestamp variables are converted to nanosec (integers)
  • do the actual value counting based on categories and created indices
  • then convert to histograms and add to datastore
fill_histogram(idf, c)

Fill input histogram with column(s) of input dataframe.

Parameters:
  • idf – input data frame used for filling histogram
  • c (list) – histogram column(s)
finalize()

Finalize the link.

Store Histograms here, if requested.

get_all_columns(data)

Retrieve all columns / keys from input data.

Parameters:data – input data sample (pandas dataframe or dict)
Returns:list of columns
Return type:list
get_data_type(df, col)

Get data type of dataframe column.

Parameters:
  • df – input data frame
  • col (str) – column
initialize()

Initialize the link.

process_and_store()

Store (and possibly process) histogram objects.

process_columns(df)

Process columns before histogram filling.

Specifically, convert timestamp columns to integers

Parameters:df – input (pandas) data frame
Returns:output (pandas) data frame with converted timestamp columns
Return type:pandas DataFrame
var_bin_specs(c, idx=0)

Determine bin_specs to use for variable c.

Parameters:
  • c (list) – list of variables, or string variable
  • idx (int) – index of the variable in c, for which to return the bin specs. default is 0.
Returns:

selected bin_specs of variable
eskapade.analysis.histogram_filling.only_bool(val)

Pass input value or array only if it is a bool.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:np.bool or np.ndarray
eskapade.analysis.histogram_filling.only_float(val)

Pass input val value or array only if it is a float.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:np.float64 or np.ndarray
eskapade.analysis.histogram_filling.only_int(val)

Pass input val value or array only if it is an integer.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:np.int64 or np.ndarray
eskapade.analysis.histogram_filling.only_str(val)

Pass input value or array only if it is a string.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:str or np.ndarray
eskapade.analysis.histogram_filling.to_ns(x)

Convert input timestamps to nanoseconds (integers).

Parameters:x – value to be converted
Returns:converted value
Return type:int
eskapade.analysis.histogram_filling.to_str(val)

Convert input to (array of) string(s).

Parameters:val – value to be converted
Returns:converted value
Return type:str or np.ndarray
eskapade.analysis.histogram_filling.value_to_bin_center(val, **kwargs)

Convert value to bin center.

Convert a numeric or timestamp column to a common bin center value.

Parameters:
  • bin_width – bin_width value needed to convert column to a common bin center value
  • bin_offset – bin_offset value needed to convert column to a common bin center value
eskapade.analysis.histogram_filling.value_to_bin_index(val, **kwargs)

Convert value to bin index.

Convert a numeric or timestamp column to an integer bin index.

Parameters:
  • bin_width – bin_width value needed to convert column to an integer bin index
  • bin_offset – bin_offset value needed to convert column to an integer bin index
eskapade.analysis.statistics module

Project: Eskapade - A python-based package for data analysis.

Classes: ArrayStats, GroupByStats

Created: 2017/03/21

Description:
Summary of an array.
Authors:
KPMG Advanced Analytics & Big Data team, Amstelveen, The Netherlands

Redistribution and use in source and binary forms, with or without modification, are permitted according to the terms listed in the file LICENSE.

class eskapade.analysis.statistics.ArrayStats(data, col_name, weights=None, unit='', label='')

Bases: object

Create summary of an array.

Class to calculate statistics (mean, standard deviation, percentiles, etc.) and create a histogram of values in an array. The statistics can be returned as values in a dictionary, a printable string, or as a LaTeX string.

__init__(data, col_name, weights=None, unit='', label='')

Initialize for a single column in data frame.

Parameters:
  • data ((keys of) dict) – Input array
  • col_name – column name
  • weights (string (column of data)) – Input array (default None)
  • unit – Unit of column
  • label (str) – Label to describe column variable
Raises:

TypeError
create_mpv_stat()

Compute most probable value from histogram.

This function computes the most probable value based on the histogram from make_histogram(), and adds it to the statistics.

create_stats()

Compute statistical properties of column variable.

This function computes the statistical properties of values in the specified column. It is called by other functions that use the resulting figures to create a statistical overview.

get_col_props()

Get column properties.

Returns dict:Column properties
get_latex_table(get_stats=None, latex=True)

Get LaTeX code string for table of stats values.

Parameters:
  • get_stats (list) – List of statistics that you want to filter on. (default None (all stats)) Available stats are: ‘count’, ‘filled’, ‘distinct’, ‘mean’, ‘std’, ‘min’, ‘max’, ‘p05’, ‘p16’, ‘p50’, ‘p84’, ‘p95’, ‘p99’
  • latex (bool) – LaTeX output or list output (default True)
Returns str:

LaTeX code snippet
get_print_stats(to_output=False)

Get statistics in printable form.

Parameters:to_output (bool) – Print statistics to output stream?
Returns str:Printable statistics string
get_x_label()

Get x label.

logger

A logger that emits log messages to an observer.

The logger can be instantiated as a module or class attribute, e.g.

>>> logger = Logger()
>>> logger.info("I'm a module logger attribute.")
>>>
>>> class Point(object):
>>>     logger = Logger()
>>>
>>>     def __init__(self, x = 0.0, y = 0.0):
>>>         Point.logger.debug('Initializing {point} with x = {x}  y = {y}', point=Point, x=x, y=y)
>>>         self._x = x
>>>         self._y = y
>>>
>>>     @property
>>>     def x(self):
>>>         self.logger.debug('Getting property x = {point._x}', point=self)
>>>         return self._x
>>>
>>>     @x.setter
>>>     def x(self, x):
>>>         self.logger.debug('Setting property y = {point._x}', point=self)
>>>         self._x = x
>>>
>>>     @property
>>>     def y(self):
>>>        self.logger.debug('Getting property y = {point._y}', point=self)
>>>        return self._y
>>>
>>>     @y.setter
>>>     def y(self, y):
>>>         self.logger.debug('Setting property y = {point._y}', point=self)
>>>         self._y = y
>>>
>>> a_point = Point(1, 2)
>>>
>>> logger.info('p_x = {point.x} p_y = {point.y}', point=a_point)
>>> logger.log_level = LogLevel.DEBUG
>>> logger.info('p_x = {point.x} p_y = {point.y}', point=a_point)

The logger uses PEP-3101 (Advanced String Formatting) with named placeholders, see <https://www.python.org/dev/peps/pep-3101/> and <https://pyformat.info/> for more details and examples.

Furthermore, logging events are only formatted and evaluated for logging levels that are enabled. So, there’s no need to check the logging level before logging. It’s also efficient.

make_histogram(var_bins=30, var_range=None, bin_edges=None, create_mpv_stat=True)

Create histogram of column values.

Parameters:
  • var_bins (int) – Number of histogram bins
  • var_range (tuple) – Range of histogram variable
  • bin_edges (list) – predefined bin edges to use for histogram. Overrules var_bins.
class eskapade.analysis.statistics.GroupByStats(data, col_name, groupby=None, weights=None, unit='', label='')

Bases: eskapade.analysis.statistics.ArrayStats

Create summary of an array in groups.

__init__(data, col_name, groupby=None, weights=None, unit='', label='')

Initialize for a single column in dataframe.

Parameters:
  • data ((keys of) dict) – Input array
  • col_name – column name
  • weights (string (column of data)) – Input array (default None)
  • unit – Unit of column
  • label (str) – Label to describe column variable
  • groupby – column name
Raises:

TypeError
get_latex_table(get_stats=None)

Get LaTeX code string for group-by table of stats values.

Parameters:get_stats (list) – same as ArrayStats.get_latex_table get_stats key word.
Returns str:LaTeX code snippet
eskapade.analysis.statistics.get_col_props(var_type)

Get column properties.

Returns dict:Column properties
eskapade.analysis.statistics.weighted_quantile(data, weights=None, probability=0.5)

Compute the weighted quantile of a 1D numpy array.

Weighted quantiles, inspired by: https://github.com/nudomarinero/wquantiles/blob/master/wquantiles.py written by Jose Sabater Here updated to return multiple quantiles in one go. Now also works when weight is None.

Parameters:
  • data (ndarray) – input array (one dimension).
  • weights (ndarray) – array with the weights of the same size of data.
  • probability (ndarray) – array of quantiles to compute. Each probablity must have a value between 0 and 1.
Returns:

list of the output value(s).
Module contents
eskapade.core package
Submodules
eskapade.core.definitions module
eskapade.core.element module
eskapade.core.exceptions module
eskapade.core.execution module
eskapade.core.meta module
eskapade.core.mixin module
eskapade.core.persistence module
eskapade.core.process_manager module
eskapade.core.process_services module
eskapade.core.run_utils module
Module contents
eskapade.core_ops package
Subpackages
Module contents
eskapade.data_quality package
Subpackages
Submodules
eskapade.data_quality.dq_helper module

Project: Eskapade - A Python-based package for data analysis.

Module: data_quality.dq_helper

Created: 2017/04/11

Description:
Data-quality helper functions
Authors:
KPMG Advanced Analytics & Big Data team, Amstelveen, The Netherlands

Redistribution and use in source and binary forms, with or without modification, are permitted according to the terms listed in the file LICENSE.

eskapade.data_quality.dq_helper.bool_to_int(val, **kwargs)

Convert input boolean to int.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:np.int64
eskapade.data_quality.dq_helper.bool_to_str(val, **kwargs)

Convert input boolean to str.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:str
eskapade.data_quality.dq_helper.check_nan(val)

Check input value for not a number.

Parameters:val – value to be checked for nan
Returns:true if nan
Return type:bool
eskapade.data_quality.dq_helper.cleanup_string(col)

Cleanup input string.

Parameters:col – string to be cleaned up
Returns:cleaned up string
Return type:str
eskapade.data_quality.dq_helper.convert(val)

Convert input to interpreted data type.

Parameters:val – value to be interpreted
Returns:interpreted value
eskapade.data_quality.dq_helper.to_date_time(val, **kwargs)

Convert input to numpy.datetime64.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:numpy.datetime64
eskapade.data_quality.dq_helper.to_float(val, **kwargs)

Convert input to float.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:np.float64
eskapade.data_quality.dq_helper.to_int(val, **kwargs)

Convert input to int.

Parameters:val – value to be evaluated
Returns:evaluated value
Return type:np.int64
eskapade.data_quality.dq_helper.to_str(val, **kwargs)

Convert input to string.

Parameters:val – value to be converted
Returns:converted value
Return type:str
Module contents
eskapade.logger package
Module contents
eskapade.visualization package
Subpackages
Submodules
eskapade.visualization.vis_utils module
Module contents
Submodules
eskapade.entry_points module
eskapade.exceptions module
eskapade.helpers module
eskapade.resources module
eskapade.utils module
eskapade.version module

THIS FILE IS AUTO-GENERATED BY ESKAPADE SETUP.PY.

Module contents

Appendices

Miscellaneous

Collection of miscelleneous Eskapade related items.

  • See Migration Tips to migrate between Eskapade versions.
  • See macOS to get started with Eskapade on a mac.
Migration Tips
From version 0.8 to 0.9

In Eskapade v0.9 the core functionality has been migrated to the separate package Eskapade-Core. We have tried to make this transition as seamless as possible, but you may well run into any migration issues. In case you do below we list the changes needed to migrate from Eskapade version 0.8 to version 0.9.

  • Whenever a line with import eskapade fails, simply replace eskapade with escore.
  • For example: from eskapade import core change to from escore import core

That’s it.

From version 0.6 to 0.7

Below we list the API changes needed to migrate from Eskapade version 0.6 to version 0.7.

Macros

  • Process manager definition:

    • proc_mgr. change to process_manager.
    • ProcessManager change to process_manager
    • Delete line: proc_mgr = ProcessManager()

  • Logger:

    • import logging change to from eskapade.logger import Logger, LogLevel
    • log. change to logger.
    • log = logging.getLogger('macro.cpf_analysis') change to logger = Logger()
    • logging change to LogLevel

  • Settings:

    Remove os.environ['WORKDIRROOT'], since the environment variable WORKDIRROOT is no longer defined, define explicitly the data and macro paths, or execute the macros and tests from the root directory of the project, resulting in something like:

    • settings['resultsDir'] = os.getcwd() + 'es_results'
    • settings['macrosDir']  = os.getcwd() + 'es_macros'
    • settings['dataDir']    = os.getcwd() + 'data'

  • Chain definition in macros:

    • To import the Chain object add from eskapade import Chain
    • Change process_manager.add_chain('chain_name') to <chain_name> = Chain('chain_name')
    • process_manager.get_chain('ReadCSV').add_link to <chain_name>.add
Tests

  • Process manager definition:

    • Change ProcessManager() to process_manager
    • Change process_manager.get_chain to process_manager.get

  • Settings:

    Remove os.environ['WORKDIRROOT'], since the environment variable WORKDIRROOT is no longer defined, define explicitly the data and macro paths, or execute the macros and tests from the root directory of the project, resulting in something like:

    • settings['macrosDir'] = os.getcwd() + '/es_macros'
    • settings['dataDir']   = os.getcwd() + '/data'

  • StatusCode:

    • Change status.isSkipChain() to status.is_skip_chain()
macOS

To install Eskapade on macOS there are basically four steps:

  • Setting up Python 3.6
  • Setting up Apache Spark 2.x
  • Setting up ROOT 6.10.08
  • Setting up Eskapade

Note

This installation guide is written for macOS High Sierra with Homebrew, and fish.

Setting up Python 3.6

Homebrew provides Python 3.6 for macOS:

$ brew install python3

To create an isolated Python installation use virtualenv:

$ virtualenv venv/eskapade --python=python3 --system-site-packages

Each time a new terminal is started, set up the virtual python environment:

$ . ~/venv/eskapade/bin/activate.fish
Setting up ROOT 6

Clone ROOT from the git repository:

git clone http://root.cern.ch/git/root.git
cd root
git checkout -b v6-10-08 v6-10-08

Then compile it with the additional flags to ensure the desired functionality:

$ mkdir ~/root_v06-10-08_p36m && cd ~/root_v06-10-08_p36m
$ cmake -Dfftw3=ON -Dmathmore=ON -Dminuit2=ON -Droofit=ON -Dtmva=ON -Dsoversion=ON -Dthread=ON -Dpython3=ON -DPYTHON_EXECUTABLE=/usr/local/opt/python3/Frameworks/Python.framework/Versions/3.6/bin/python3.6m -DPYTHON_INCLUDE_DIR=/usr/local/opt/python3/Frameworks/Python.framework/Versions/3.6/include/python3.6m/ -DPYTHON_LIBRARY=/usr/local/opt/python3/Frameworks/Python.framework/Versions/3.6/lib/libpython3.6m.dylib $HOME/root
$ cmake --build . -- -j7

PS: make sure all the flags are picked up correctly (for example, -Dfftw3 requires fftw to be installed with Homebrew).

To setup the ROOT environment each time a new shell is started, set the following environment variables:

set -xg ROOTSYS "$HOME/root_v06-10-08_p36m"
set -xg PATH $ROOTSYS/bin $PATH
set -xg LD_LIBRARY_PATH "$ROOTSYS/lib:$LD_LIBRARY_PATH"
set -xg DYLD_LIBRARY_PATH "$ROOTSYS/lib:$DYLD_LIBRARY_PATH"
set -xg LIBPATH "$ROOTSYS/lib:$LIBPATH"
set -xg SHLIB_PATH "$ROOTSYS/lib:$SHLIB_PATH"
set -xg PYTHONPATH "$ROOTSYS/lib:$PYTHONPATH"

Note that for bash shells this can be done by sourcing the script in root_v06-10-08_p36m/bin/thisroot.sh.

Finally, install the Python packages for ROOT bindings:

$ pip install rootpy==1.0.1 root-numpy=4.7.3
Setting up Apache Spark 2.x

Apache Spark is provided through Homebrew:

$ brew install apache-spark

The py4j package is needed to support access to Java objects from Python:

$ pip install py4j==0.10.4

To set up the Spark environment each time a new terminal is started set:

set -xg SPARK_HOME (brew --prefix apache-spark)/libexec
set -xg SPARK_LOCAL_HOSTNAME "localhost"
set -xg PYTHONPATH "$SPARK_HOME/python:$PYTHONPATH"
Setting up Eskapade

To set up the Eskapade environment (Python, Spark, ROOT) each time a new terminal is started, source a shell script (e.g. setup_eskapade.fish) that contains set the environment variables as described above:

# --- setup Python
. ~/venv/eskapade/bin/activate.fish

# --- setup ROOT
set -xg ROOTSYS "${HOME}/root_v06-10-08_p36m"
set -xg PATH $ROOTSYS/bin $PATH
set -xg LD_LIBRARY_PATH "$ROOTSYS/lib:$LD_LIBRARY_PATH"
set -xg DYLD_LIBRARY_PATH "$ROOTSYS/lib:$DYLD_LIBRARY_PATH"
set -xg LIBPATH "$ROOTSYS/lib:$LIBPATH"
set -xg SHLIB_PATH "$ROOTSYS/lib:$SHLIB_PATH"
set -xg PYTHONPATH "$ROOTSYS/lib:$PYTHONPATH"

# --- setup Spark
set -xg SPARK_HOME (brew --prefix apache-spark)/libexec
set -xg SPARK_LOCAL_HOSTNAME "localhost"
set -xg PYTHONPATH "$SPARK_HOME/python:$PYTHONPATH"

# --- setup Eskapade
cd /path/to/eskapade

Finally, install Eskapade (and it’s dependencies) by simply running:

$ pip install Eskapade

Indices and tables