Semi-Supervised Meta-Learning

This repository contains code for my project on Semi-Supervised Meta Learning for Stanford class CS330.

The projet report, containing a detailed description of the project and its results is here.

The implementation of Constrained DeepCluster is here: Constrained DeepCluster

Usage

In order to run the experiments as described in the project report, go through the following steps:

1. Follow the instructions in the Constrained DeepCluster repository to download and prepare the mini Imagenet data set and run both DeepCluster and Constrained DeepCluster.
2. The following files are created as a result of running DeepCluster and Constrained DeepCluster: labeled_tasks.npy, embedding.npy, images.npy, embedding_standard_labeled.npy,embedding_standard_unlabeled.npy,embedding_labeled.npy, embedding_unlabeled.npy, images_unlabeled.npy. Copy these files over here into this directory.
3. Run make_clusterings.sh in order to create partitions by using k-means and constrained k-means on partitions.
4. Run proto_experiments.sh in order to run the experiments with ProtoNets and/or run maml_experiments.sh in order to run the experiments with MAML.

Requirements

• Python 3.7
• tensorflow 2
• scikit-learn 0.23.2