GriP on Data Science

Based on Cookiecutter Data Science, this Data Science project template aims to provide a more opinionated and firm framework, especially tailored towards beginners.

Why use this project structure?

While the Cookiecutter Data Science template aims at being a great starting point, it leaves you with a lot of design choices:

Certainly, one could argue that the answer to these questions depends on the concrete project at hand. However, I think that beginning Data Scientists especially (or rather beginners in any field, really) need a firm and more refined framework to start their journey. To be honest, when I found out about the Data Science cookiecutter, I was overwhelmed by the potential and the choices I had to make when starting my first Data Science project based on the template.

Furthermore, there are a lot of forks of the Cookiecutter Data Science template, a lot of which aim at providing more orientation. The main example I was inspired by is Cookiecutter EasyData that provides a plethora of additional make targets, is able to manage dependencies via conda, and adds a lot of boilerplate code assisting in building up a Data Science project.

While the aforementioned template tackles a lot of the weaknesses of the Cookiecutter Data Science template, I find that it is too sophisticated and not well-documented enough to be easy to use (even though the maintainers started an example project explaining their template structure and how to use it). Thus, I decided to take the middle ground between the two.

Start with 100% test coverage

Depending on the amount of data, transformations may take quite some time and should be tested before they run on the whole dataset -- it's aggravating when a ten minute transformation yields the wrong result because you applied it to the wrong column.

This project provides you with tests that illustrate how your command line scripts that apply the transformation can be tested.

Be as elaborate as you want with your log messages

With the logging configuration that comes with this project, all INFO level messages will be emitted to the console. If you need more detailed information but do not want to pollute the console, use the DEBUG level that will be automatically logged into log files in the log folder.

Code style: black

The initial code and tests are formatted by black.

Use pre-defined make targets for common tasks

There are pre-defined make targets that

The latter are configured via simple dictionaries. Details below.

Getting started

Requirements

$ pip install cookiecutter

or

$ conda config --add channels conda-forge
$ conda install cookiecutter

Starting a new project

Starting a new project is as easy as running this command at the command line. No need to create a directory first, the cookiecutter will do it for you.

cookiecutter https://github.com/waveFrontSet/cookiecutter-data-science

Create a new environment

After the new project has been created you should create a new conda environment using

make create_environment

Activate it using conda activate project_name.

Implement the data/raw target

The data/raw target is the specification of how to obtain the raw data of your project. At the beginning, it is left empty. Decide what you need to do to obtain the raw data and decide how to save it. If necessary, write a script next to the generic_processing.py script.

Implement generic clean ups and transformations

The data/interim target is for generic clean ups and transformations (examples: Removing unnecessary columns, removing duplicated rows). It aims at the generic_processing.py script. Edit it according to your needs.

Implement transformations and aggregations resulting in your final dataset

The data/processed/dataset.csv target is the road to your final dataset. This dataset will be used to train a model. The target aims at the build_features.py script. Edit it according to your needs.

Decide which models to train

Models are trained via the train target. Edit the MODELS dictionary in the model_config.py file to configure which models to train. The TARGET_VALUE constant should contain the name of the column that is the prediction target of your models.

Example: If you want to fit a Linear Regression model as well as a Decision Tree with standard hyperparameters, use

from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor

MODELS = {
    "linearRegression": LinearRegression(),
    "decisionTree": DecisionTreeRegressor(),
}

Issuing make train will fit both models to the train set and dump the resulting models into the folder models/. The filename of each model will be its key.

Note that this target automatically invokes the train_test_split target that splits data/processed/dataset.csv into a train set train.csv and a test set test.csv.

Evaluate your models

Models are evaluated via the evaluate target. Edit the METRICS dictionary in the metric_config.py file to configure which metrics to evaluate for each model.

Example: If you want to evaluate both models from the above paragraph using R2-Score and mean absolute error, use

from sklearn.metrics import r2_score, mean_absolute_error

METRICS = {
    "linearRegression": [r2_score, mean_absolute_error],
    "decisionTree": [r2_score, mean_absolute_error],
}

Issuing make evaluate will deserialize both models and evaluate them using the defined metrics on the test set. The results are printed to the console and, additionally, will be written into the file models/report.json.

Directory structure

The majority is the same as in the basic Cookiecutter Data Science template. Added are the tests folder as well as the model_config.py and the metric_config.py. Plus, make_dataset.py was renamed to generic_processing.py.

├── LICENSE
├── Makefile           <- Makefile with commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- A default Sphinx project; see sphinx-doc.org for details
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.py           <- Make this project pip installable with `pip install -e`
├── {{ cookiecutter.module_name }}                <- Source code for use in this project.
│   ├── __init__.py    <- Makes {{ cookiecutter.module_name }} a Python module
│   │
│   ├── data           <- Scripts to download or generate data and for generic transformations
│   │   └── generic_processing.py
│   │
│   ├── features       <- Scripts to turn raw data into features for modeling
│   │   └── build_features.py
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions
│   │   ├── metric_config.py <- Contains the dictionary that configures the
│   │   │                       metrics used to evaluate your models
│   │   ├── model_config.py <- Contains the dictionary that configures the
│   │   │                       models to train and the target value
│   │   ├── predict_model.py
│   │   └── train_model.py
│   │
│   └── visualization  <- Scripts to create exploratory and results oriented visualizations
│       └── visualize.py
├── tests              <- Folder with initial tests for 100% coverage
└── tox.ini            <- tox file with settings for running tox; see tox.testrun.org

Keep secrets and configuration out of version control

You really don't want to leak your AWS secret key or Postgres username and password on Github. Enough said — see the Twelve Factor App principles on this point. Here's one way to do this:

Store your secrets and config variables in a special file

Create a .env file in the project root folder. Thanks to the .gitignore, this file should never get committed into the version control repository. Here's an example:

# example .env file
DATABASE_URL=postgres://username:password@localhost:5432/dbname
AWS_ACCESS_KEY=myaccesskey
AWS_SECRET_ACCESS_KEY=mysecretkey
OTHER_VARIABLE=something

Use a package to load these variables automatically.

If you look at the stub script in {{ cookiecutter.module_name }}/data/make_dataset.py, it uses a package called python-dotenv to load up all the entries in this file as environment variables so they are accessible with os.environ.get. Here's an example snippet adapted from the python-dotenv documentation:

# {{ cookiecutter.module_name }}/data/dotenv_example.py
import os
from dotenv import load_dotenv, find_dotenv

# find .env automagically by walking up directories until it's found
dotenv_path = find_dotenv()

# load up the entries as environment variables
load_dotenv(dotenv_path)

database_url = os.environ.get("DATABASE_URL")
other_variable = os.environ.get("OTHER_VARIABLE")

AWS CLI configuration

When using Amazon S3 to store data, a simple method of managing AWS access is to set your access keys to environment variables. However, managing mutiple sets of keys on a single machine (e.g. when working on multiple projects) it is best to use a credentials file, typically located in ~/.aws/credentials. A typical file might look like:

[default]
aws_access_key_id=myaccesskey
aws_secret_access_key=mysecretkey

[another_project]
aws_access_key_id=myprojectaccesskey
aws_secret_access_key=myprojectsecretkey

You can add the profile name when initialising a project; assuming no applicable environment variables are set, the profile credentials will be used be default.