1. Clone the repository

  $ git clone https://github.com/RahulBaboota/Semi-Supervised-Clustering-using-Neural-Networks.git
  $ cd Semi-Supervised-Clustering-using-Neural-Networks

2. Create new virtual environment

  $ sudo pip install virtualenv
  $ virtualenv venv
  $ source venv/bin/activate
  $ pip install -r requirements.txt

3. Repository Information

• The file Test.py contains the code for testing the auto-encoder.
• The file DataLoader.py contains the different data loader functions.
• The file PreTraining.py contains the code for pretraining the auto-encoder.
• The file Visualisations.py contains the code for producing the plots present in the report.
• The file Train.py contains the code for training the auto-encoder on the custom loss function.
• The file Model.py contains the architecture definition of the convolutional auto-encoder that is used.

4. Training the auto-encoder network

  $ python Train.py --dataSet <dataSet Name> --percentLabData <Percentage of labelled data to be used>

The training procedure will train the convolutional auto-encoder on the specified dataset. The folder /TrainingResults will get populated with the trained model as well as various other information required for generating the plots.

5. Testing the auto-encoder network

  $ python Test.py --dataSet <dataSet Name> --percentLabData <Percentage of labelled data to be used>

The testing script will run the trained model on the test data and will populate the folder /TestingResults .

5. Producing the plots

  $ python Visualisations.py --dataSet <dataSet Name> --percentLabData <Percentage of labelled data to be used>

This script will produce the different plots for the dataset and percentage of labelled data specified in the input.

6. Example Plots

The plots generated for 10% labelled data for MNIST are displayed below.