m0d3l

---

Description

Build, train and test Neural Net Models based using Pytorch. This package builds NN models from the artefacts created with the f3atur3s and eng1n3 packages.

See the notebooks directory for examples.

Example usage

# Define the Model
class FirstModel(mp.BinaryClassifier):
    def __init__(self, model_configuration: mp.ModelConfiguration):
        super(FirstModel, self).__init__(model_configuration)
        self.heads = self.create_heads()
        head_size = sum([h.output_size for h in self.heads])
        self.tail = self.create_tail(head_size)

    def forward(self, x: Tuple[torch.Tensor, ...]) -> Tuple[torch.Tensor,...]:
        o = torch.cat([h(x[i]) for i, h in enumerate(self.heads)], dim=1)
        o = self.tail(o)
        return (o,)

# Create a Model
model = FirstModel(mp.ModelConfiguration.from_tensor_definitions(ti.target_tensor_def))

# Create a trainer.
trainer = mp.Trainer(model, torch.device('cpu'), train_dl, val_dl)
# And an optimizer
optimizer = mp.AdamWOptimizer(model, lr=0.01)
# Run the trainer for 5 epochs
history = trainer.train(5, optimizer)

Requirements

• pandas
• numpy
• numba
• torch
• tqdm
• matplotlib
• scikit-learn