training
Functions
|
Begin training the Unet model. |
|
Prepare the input data for the model. |
Module Contents
- training.begin_training(savedir, stage, xtrain, ytrain, xvalid, yvalid, unet, batch_size=30, n_epochs=250, save_format='keras')
Begin training the Unet model.
- Parameters:
savedir (str) – Directory to save outputs.
stage (int) – The stage number (1 or 2).
xtrain (np.ndarray) – Training input features.
ytrain (np.ndarray) – Training target variables.
xvalid (np.ndarray) – Validation input features.
yvalid (np.ndarray) – Validation target variables.
unet (Unet) – The Unet model to be trained.
batch_size (int, optional) – Batch size for training.
n_epochs (int, optional) – Number of epochs for training.
save_format (str, optional) – Format to save the model (‘h5’, ‘keras’, or ‘both’).
- Returns:
unet – The trained Unet model.
- Return type:
- training.make_predictions(uarr, unet, config_dict, config_path, output_metadata, stage=1)
Prepare the input data for the model.
Get the training data from the input NetCDF dataset as numpy arrays and concatenate them along the time dimension.
- Parameters:
uarr (unox.uarray) – The dataset of the input NetCDF file.
unet (Unet) – The Unet model to be trained.
config_dict (dict) – A dictionary containing the configuration.
config_path (str) – Path to the input configuration JSON file used to make config_dict.
output_metadata (dict) – The dictionary of metadata describing the output of a model run.
stage (int) – The stage of the data to plot (1 or 2).
- Returns:
xtrain (np.ndarray) – Concatenated training input features.
ytrain (np.ndarray) – Concatenated training target variables.
output_metadata (dict) – The dictionary of metadata describing the output of a model run with values added for train_years and unet_build_shape.