This assignment is about applying a convolutional neural network (CNN) to the task of digit recognition using the PyTorch framework. Students are required to improve the recognition accuracy of this CNN by tuning several hyper-parameters and submit a report that summarizes their trials.
We will use different optimizer to try to find the best settings and tune the parameters for each optimizer, during which we will use the control variate method. Also, we may change the net architecture and the other parameters like the convolution kernel size.
The zip file "data.zip" containing the training and testing data can be downloaded from this link: https://drive.google.com/file/d/1_kqIftWGqKZCCmD9-2WQKHGgLgNyufRt/view?usp=sharing
Download the zip file, unzip it to get the "data" folder. Put the "data" folder and "final.py" into the same folder so that the code can run correctly.
The python libraries we used are sklearn, torch, torchvision and numpy.