visenger / ml-powered-applications

Companion repository for the book Building Machine Learning Powered Applications

Building ML Powered Applications

Welcome to the companion code repository for the O'Reilly book Building ML Powered Applications. The book is available on Amazon.

This repository consists of three parts:

• A set of Jupyter notebooks in the notebook folder serve to illustrate concepts covered in the book.

• A library in the ml_editor folder contains core functions for the book's case study example, a Machine Learning driven writing assistant.

• A Flask app demonstrates a simple way to serve results to users

Credit and thanks go to Bruno Guisard who conducted a thorough review of the code in this repository.

Setup instructions

Python environment

This repository has been tested on Python 3.6 and 3.7. It aims to support any Python 3 version.

To setup, start by cloning the repository:

git clone https://github.com/hundredblocks/ml-powered-applications.git

Then, navigate to the repository and create a python virtual environment using virtualenv:

cd ml-powered-applications

virtualenv ml_editor

You can then activate it by running:

source ml_editor/bin/activate

Then, install project requirements by using:

pip install -r requirements.txt

The library uses a few models from spacy. To download the small and large English model (required to run the app and the notebooks), run these commands from a terminal with your virtualenv activated:

python -m spacy download en_core_web_sm

python -m spacy download en_core_web_lg

Finally, the notebooks and library leverage the nltk package. The package comes with a set of resources that need to be individually downloaded. To do so, open a Python session in an activated virtual environment, import nltk, and download the required resource.

Here is an example of how to do this for the punkt package from an active virtual environment with nltk installed:

python

import nltk

nltk.download('punkt')

Notebook examples

The notebook folder contains usage examples for concepts covered in the book. Most of the examples only use one of the subfolders in archive (the one that contains data for writers.stackexchange.com).

I've included a processed version of the data as a .csv for convenience.

If you wanted to generate this data yourself, or generate it for another subfolder, you should:

• Download a subfolder from the stackoverflow archives

• Run parse_xml_to_csv to convert it to a DataFrame

• Run generate_model_text_features to generate a DataFrames with precomputed features

The notebooks belong to a few categories of concepts, described below.

Data Exploration and Transformation

• Dataset Exploration
• Splitting Data
• Vectorizing Text
• Clustering Data
• Tabular Data Vectorization
• Exploring Data To Generate Features

Initial Model Training and Performance Analysis

• Train Simple Model
• Comparing Data To Predictions
• Top K
• Feature Importance
• Black Box Explainer

Improving the Model

• Second Model
• Third Model

Model Comparison

• Comparing Models

Generating Suggestions from Models

• Generating Recommendations

Pretrained models

You can train and save models using the notebooks in the notebook folder. For convenience, I've included three trained models and two vectorizers, serialized in the models folder. These models are loaded by notebooks demonstrating methods to compare model results, as well as in the flask app.

Running the prototype Flask app

To run the app, simply navigate to the root of the repository and run:

FLASK_APP=app.py flask run

The above command should spin up a local web-app you can access at http://127.0.0.1:5000/

Troubleshooting

If you have any questions or encounter any roadblocks, please feel free to open an issue or email me at [e-mail].

Project structure inspired by the great Cookiecutter Data Science.