The breizhcrops from maxzoll

maxzoll / breizhcrops Goto Github PK

View Code? Open in Web Editor NEW

A Satellite Time Series Dataset for Crop Type Identification

License: GNU General Public License v3.0

Jupyter Notebook 61.11% Python 38.68% Shell 0.14% Dockerfile 0.07%

breizhcrops's People

Contributors

breizhcrops's Issues

New models

Hi Max,
I just added the inceptiontime model in the upstream repo

hyperparameter are hidden_dims and num_layers
see

        model = InceptionTime(input_dim=ndims, num_classes=num_classes,
                              num_layers=args.hyperparameter["num_layers"],
                              hidden_dims=args.hyperparameter["hidden_dims"], device=device).to(device)

Also, I replaced the transformer implementation with the new one provided by pytorch (makes the code much shorter!)

The hyperparameters are: hidden_dims, n_head, num_layers,relu,dropout
see

        model = Transformer(input_dim=ndims, num_classes=num_classes, sequencelength=sequencelength,
                            d_model=args.hyperparameter["hidden_dims"],
                            n_head=args.hyperparameter["n_head"],
                            n_layers=args.hyperparameter["num_layers"],
                            d_inner=4*args.hyperparameter["hidden_dims"],
                            activation="relu",
                            dropout=args.hyperparameter["dropout"]
                            ).to(device)

I developed the code until training started on my side. Do you think you could add these to the hyperparameter tuning as well?

Recommend Projects