Simple-Two-Layer-Neural-Network

The code is from CMU 10714(Deep Learning System)'s homework, and I have attached the homework description(hw0.iqynb and manual_neural_nets.pdf). Also, can refer to my DeepLearningSystem repository for whole description. Also can refer to 🌟🌟🌟Wenqing machine learning blog's backpropagation.md and ffnn.md🌟🌟🌟.

It implemented softmax regression which using softmax(aka cross entropy) loss on the MNIST dataset.

Homework Description

• hw0.iqynb from question 2, 3 and 5

How to run the code

Training softmax regression:

python3 src/simple_ml.py  

Testing all functions:

python3 -m pytest tests/test_simple_ml.py       

Testing a single function:

 python3 -m pytest -k "softmax_loss" tests/test_simple_ml.py

The -k option in pytest is used to filter tests based on the names of the test functions, test class names, and test file names. It uses a substring match to select tests. When you specify -k "softmax_regression_epoch and not cpp", pytest is selecting tests that match the following criteria:

1. Contain softmax_regression_epoch: Any test function, test class, or test file that contains the substring softmax_regression_epoch.
2. Do not contain cpp: Any test function, test class, or test file that does not contain the substring cpp.

This filter is a logical combination, where the and operator means that both conditions must be satisfied, and the not operator negates the second condition.