snakepot

snakepot is a snakemake workflow designed to train and evaluate a binary classifier using the TPOT auto-ML library.

Quick Start

1. Install snakepot (requires conda)

2. Edit parameters in config.json

3. Run snakemake

4. View outputs in new directory

Features

I developed snakepot during my elective at the William Harvey Research Institute. We used snakepot to quickly train a baseline model on a variety of gene-phenotype datasets. The workflow takes the following steps:

1. Clean the data set (simple N/A drop by rows and explicit drop by columns)
2. Split data into train/test/validate sets
3. Call the TPOT automated machine learning algorithm to train a classifier
4. Save classifier and re-run it on the houldout/validation data
5. Evaluate the classifier on the holdout set
6. Call the classifier for predictions on the unlabelled data

An example dataset (/test/data.csv) and config file (config.json) are provided.

Setup

# Build conda environment
conda env create -f environment.yaml
conda activate snakepot
# Install python helper scripts to path
pip install . 
# Run the workflow in Snakefile using config.json
snakemake 

config.json

Parameter	Description
directory	Output directory for new files
input	Input CSV file. All data must be encoded as binary or continuous variables
drop_columns	Features to drop from the data. Skipped if not found
encode_columns	Categorical features to encode. Skipped if not found
target_column	The name of the target variable
target_1	Target variable value to label as '1'
target_0	Target variable value to label as '0'
to_predict	Target variable value for final predictions
perc_split	Percentage of training data (target '1' or '0') to split for holdout set
TPOT_max_time	Maximum time to run TPOT in muntes

License

MIT License. Copyright (c) 2019 Nana Mensah