Automatic Machine Learning solver

A maching learning problem solver for typical kaggle datasets. (ie. with a trainset and a testset)

Table of contents

• General info
• Technologies
• Features
• Run
• Status
• Inspiration
• Contact

General info

The Automatic Machine Learning solver has the goal to give you quick insight about your Machine Learning problem. It's give you 3 "clean" quick and dirty models, and them benchmarks scores. With that insights, you can have a best estimation of the performance you can achieve, and what kind of models work better. (Linear, boosting, or bagging)

Screenshots

Technologies

• python - Version 3.7
• pandas - Version 0.25
• scikit-learn - Version 0.21

The code was tested and adapted for these dependencies.

Features

List of features ready

• Missing Values handling
• Categorial features handling for each kind of model
• Optimization of each model by GridSearch
• Model evaluation (train and valid set)

To-do list:

• GridSearch CV is optimized by the default scorer of the estimator (which can be non adapted for example in case of a classification problem with imbalanced class, the default scorer is the accuracy which should not be used)
• The solver is still very generic

Run

Go to the main.py script and precise every argument of the main function:

• output_dir: Where the result should be stored
• name: The name of the problem to solve
• type_ml: The type of Machine Learning problem: Classification 'clf' or Regression 'r'
• path_train: The path to the csv trainfile
• path_test: The path to the csv testfile
• id_col: The column of the id
• cible_col: The column of the target

Status

Project is: in progress

Contact

Created by @bernadinel - feel free to contact me!