This project aims at implementing different types of Recurrent Neural Networks (RNNs) in pure C without any external libraries.
The models are trained on the weather data of the city of New Dehli, India and predicts the temperature.

The motivation behind this project is to deeply understand how LSTMs work as I believe I will be using them in a future project but also to compare them to other models.

Make sure to have the data directory in the same directory than the executables if you are using them. In that case your folder should look like this:

├── vanilla.exe
├── lstm.exe
├── gru.exe
└── data
    ├── train.csv
    └── test.csv

To compile the Vanilla RNN run:

gcc -g vanilla.c lib/data.c lib/vanilla/model.c lib/vanilla/forward.c lib/vanilla/backprop.c -o vanilla -lm

To compile the LSTM run:

gcc -g lstm.c lib/data.c lib/lstm/model.c lib/lstm/forward.c lib/lstm/backprop.c -o lstm -lm

To compile the GRU run:

gcc -g gru.c lib/data.c lib/gru/model.c lib/gru/forward.c -o gru -lm

In order to test the models, we run them on the New Dehli weather data for 500 epochs with a learning rate of 0.001 and a hidden size of 20. The accuracy represents the percentage of predictions within 2 degrees of the actual temperature.

[Epoch 500] Loss: 4.173683
[Epoch 500] Accuracy: 67.031464%

[Test] Average Loss: 7.815743
[Test] Average Accuracy: 61.739130%

[Epoch 500] Average Loss: 2.770795
[Epoch 500] Average Accuracy: 80.095759

[Testing] Average Loss: 4.500015
[Testing] Average Accuracy: 75.438596%