oliverwong258 / solving-3sat-with-gtn Goto Github PK

This is an adjustment of the GTN model from SAT-Solver-usingNNs, primarily involving adjustments in the number of attention heads in the self-attention mechanism and modifications to the activation function of the neural network. These changes aim to enhance the model's performance, especially in scenarios where there is a significant difference in the complexity of training and testing data.

To run the program, enter python main.py --m <model_path> --s <separate> in the command line.

Here, <model_path> is the name of the saved model, and <separate> indicates whether to conduct separate testing, with 0 for no and 1 for yes.

Other optional parameters include:

• --d Path to training data, default is ./data
• --e Model embedding dimension
• --h Number of attention heads in the model(the default is 2)
• --l Number of layers in the model
• --r Dropout rate
• --ls Neuron density of the final linear layer
• --b Batch size

The output model files are saved in the ./models directory, and the learning curves are saved in the ./plots directory.