Implement a multi-layered classifier where weights of each layer is calculated greedily using layer-wise pretraining with the help of auto-encoders on STL-10 dataset. Train a classifier having X structure (excluding input and output layers) for classification task on the test set. 1. Report the classification accuracy on the test set and plot loss curves on the training and evaluation set. 2. Report the class-wise accuracy of each class. 3. Plot t-sne for this model (use embeddings from layer X[3]) . Use the first 500 images of each class from the test dataset for this visualization. X = [1024,1200,728,512,128]

Python code, that can be implemented in google colab