Compare ReLU, Sigmoid, and Tanh activation functions

Modify your PyTorch or sklearn baseline to test three different hidden layer activation functions (ReLU, Sigmoid, Tanh). Plot their training loss curves on the same graph and write a short summary of which converges the fastest and why.

Start by locating the existing PyTorch or scikit-learn baseline in the repository and run it to understand its training-loss output. Compare hidden-layer ReLU, Sigmoid, and Tanh runs on one graph, then write the requested convergence summary; the work is done when all three curves and the explanation are included.

Written by the indexing model from the issue text.