ANOVA for neural networks
Perform ANOVA (Analysis of Variance) on neural networks. We will take a basic MNIST implementation and run the experiments for different hyperparameters. Then we will try to measure the effect of different hyperparameters on performance and the interactions among the hyperparameters.
First few steps:
- Train a simple MNIST classifier and get the same results on different runs. (Use a random seed)
- Train again with different hyperparameters. (For example, learning rate and number of layers) (Atleast 10 runs each)
- Create a PR with organized results and graphs.