Introduction to Machine Learning for Text Classification

Source code for my 2022 text classification guest lecture at NHL Stenden Leeuwarden.

Prerequisites

To run this project, you'll need:

• Python 3.6 or later with pip and virtual environments

Setup

Setup will vary depending on whether you're using Windows or Mac/Linux. A script is included for each OS that will do the following for you:

• Delete any existing virtual environment for this project
• Provision fresh virtual environment
• Install all dependencies for this project in that virtual environment

Windows

On Windows, execute setup-venv.bat via your command prompt from the root directory of the repository:

./setup-venv.bat

On Mac/Linux

On Mac/Linux, execute setup-venv.sh via your terminal from the root directory of the repository. Make sure to invoke it using bash:

bash ./setup-venv.sh

Training a Model

First, prep your data. Do this like so:

1. Place each of your documents into a separate plain text (*.txt) file (check out split_csv.py for a starting point to do this programmatically)
2. Create a folder called data in the repository root directory
3. Within this data directory, make subfolders corresponding to your classes
4. Place your documents (text files) in the appropriate subfolder according to their class

Next, make sure you're in the virtual environment we set up earlier:

• On Windows - Run ./venv/Scripts/activate in your command prompt from the repo root directory
• On Mac/Linux - Tun . venv/bin/activate in your terminal from the repo root directory

Now, go ahead and run train.py like so:

python3 ./train.py

The program will train the model and print its confusion matrix, classification report and overall accuracy to the command line and plot/show the confusion matrix using matplotlib.

Your trained model will be picked to disk at ./classifier.pickle ready to load and use from your own programs.

Querying the Model

Now, you can run interact.py which will start a program that will accept typed input, run it through your trained classifier and display the classification.

Alternatively, you can load and use the model programmatically:

from joblib import load

# Load trained model.
pipeline = load('./classifier.pickle')

# Use your trained model here...
# pipeline.predict(['<My input>'])

Limitations

This example code has some limitations:

• train.py will only train models on unigrams out-of-the-box. Check out the ngram_range parameter of CountVectorizer if you're interested in training on ngrams!
• Stopwords will not be removed for you. Take a look at the stop_words parameter of CountVectorizer if you're interested in stopword removal!

This article on Practical Data Science will point you in the right direction when it comes to the above.

Cross-validation, minimization of bias etc. are very important. Wield AI/ML responsibly.