Our project intends to test different ratios of labeled to unlabeled data (R) in attempts to find the minimum ratio necessary to achieve peak accuracy, or to justify that more unlabeled data does in fact mean better performance.
"Noisy_Student_SVHN_Final_0329.ipynb" is the final code to do self-training process on SVHN dataset and generate teacher and student models.
"Noisy_Student_Mnist_R1toR6.ipynb" is the final code to do self-training process on MNIST dataset
"Predictions_and_Evaluation.ipynb" is the code to visualize evaluation metrics of different models trained with different R values
"Predictions_and_Evaluation_Mnist.ipynb" is the code to visualize evaluation metrics on MNIST dataset