ModelMaker

Framework and utilities for creating and validating ML models

Features

• CLI tool for creation and validation of models
• Choose from predefined templates

Install the tool

This will install the modelmaker CLI tool

git clone https://github.com/shirecoding/ModelMaker.git
cd ModelMaker
pip3 install ./

alternatively install from pypi

pip3 install model-maker

List Templates

modelmaker templates

Create a New Project

modelmaker new --project MNISTClassifier --package mnistmodel --template default

This will create the python package mnistmodel inside directory MNISTClassifier (also the main class name)

Current Templates

• default (simple classification model using mnist dataset example)
• linear_regression
• text_classification

Project Structure

• The model is packaged as an importable and installable python library
• scripts/train.py used for training the model and exporting it to saved_models
• scripts/test.py gives an example of how to use the packaged model in production

src/
    MNISTClassifier/
        mnistmodel/ 	# python model package
        saved_model/ 	# this is where the model is saved after training
            mnistmodel
        scripts/
            train.py 	# training script which imports model package, trains model, saves model to saved_model
            test.py 	# example test script on how to use the model package in production
        setup.py
        README.md
        .gitignore
        ...

Model Framework and Pipeline

• Each model inherits from an abstract class ModelInterface
• Each model must override get_model, fit_model, load_model, save_model
• Each model must define preprocess, predict, postprocess