icenet.model package

Submodules

icenet.model.callbacks module

class icenet.model.callbacks.BatchwiseModelCheckpoint(save_frequency: object, model_path: object, mode: object, monitor: object, prev_best: object = None, sample_at_zero: object = False)

Bases: Callback

Parameters:

• save_frequency

• model_path

• mode

• monitor

• prev_best

• sample_at_zero

on_train_batch_end(batch: object, logs: object = None)

Parameters:

• batch

• logs

class icenet.model.callbacks.IceNetPreTrainingEvaluator(*args, validation_frequency: int, val_dataloader: object, sample_at_zero: bool = False, **kwargs)

Bases: Callback

Custom tf.keras callback to update the logs dict used by all other callbacks with the validation set metrics. The callback is executed every validation_frequency batches.

This can be used in conjuction with the BatchwiseModelCheckpoint callback to perform a model checkpoint based on validation data every N batches - ensure the save_frequency input to BatchwiseModelCheckpoint is also set to validation_frequency.

Also ensure that the callbacks list past to Model.fit() contains this callback before any other callbacks that need the validation metrics.

Also use Weights and Biases to log the training and validation metrics.

TODO: not sure this is really necessary/stable, review

Parameters:

• validation_frequency

• val_dataloader

• sample_at_zero

on_train_batch_end(batch: object, logs: object = None)

Parameters:

• batch

• logs

icenet.model.losses module

class icenet.model.losses.WeightedMSE(name: str = 'mse', **kwargs)

Bases: MeanSquaredError

Custom keras loss/metric for mean squared error

Parameters:

name

icenet.model.metrics module

class icenet.model.metrics.ConstructLeadtimeAccuracy(*args, **kwargs)

Bases: CategoricalAccuracy

Computes the network’s accuracy over the active grid cell region for either a) a specific lead time in months, or b) over all lead times at once.

Parameters:

• name

• use_all_forecast_months

• single_forecast_leadtime_idx

classmethod from_config(config: object)

For saving and loading networks with this custom metric. :param config: :return:

get_config()

For saving and loading networks with this custom metric.

Returns:

result()

Returns:

update_state(y_true: object, y_pred: object, sample_weight: object = None)

Parameters:

• y_true

• y_pred

• sample_weight

class icenet.model.metrics.WeightedBinaryAccuracy(*args, **kwargs)

Bases: BinaryAccuracy

Parameters:

leadtime_idx

get_config()

For saving and loading networks with this custom metric.

Returns:

result()

Returns:

update_state(y_true: object, y_pred: object, sample_weight: object = None)

Custom keras loss/metric for binary accuracy in classifying SIC>15%

Parameters:

• y_true – Ground truth outputs

• y_pred – Network predictions

• sample_weight – Pixelwise mask weighting for metric summation

Returns:

Root mean squared error of SIC (%) (float)

class icenet.model.metrics.WeightedMAE(*args, **kwargs)

Bases: MeanAbsoluteError

Custom keras loss/metric for mean absolute error

Parameters:

• name

• leadtime_idx

result()

Returns:

update_state(y_true: object, y_pred: object, sample_weight: object = None)

Parameters:

• y_true

• y_pred

• sample_weight

Returns:

class icenet.model.metrics.WeightedMSE(*args, **kwargs)

Bases: MeanSquaredError

Custom keras loss/metric for mean squared error

Parameters:

• leadtime_idx

• name

result()

Returns:

update_state(y_true: object, y_pred: object, sample_weight: object = None)

Parameters:

• y_true

• y_pred

• sample_weight

Returns:

class icenet.model.metrics.WeightedRMSE(*args, **kwargs)

Bases: RootMeanSquaredError

Custom keras loss/metric for root mean squared error

Parameters:

• leadtime_idx

• name

result()

Returns:

update_state(y_true: object, y_pred: object, sample_weight: object = None)

Parameters:

• y_true

• y_pred

• sample_weight

Returns:

icenet.model.models module

class icenet.model.models.TemperatureScale(*args, **kwargs)

Bases: Layer

Temperature scaling layer

Implements the temperature scaling layer for probability calibration, as introduced in Guo 2017 (http://proceedings.mlr.press/v70/guo17a.html).

call(inputs: object, **kwargs)

Divide the input logits by the T value.

Parameters:

• **kwargs –

• inputs

Returns:

get_config()

For saving and loading networks with this custom layer.

Returns:

icenet.model.models.linear_trend_forecast(usable_selector: object, forecast_date: object, da: object, mask: object, missing_dates: object = (), shape: object = (432, 432)) → object

Parameters:

• usable_selector

• forecast_date

• da

• mask

• missing_dates

• shape

Returns:

icenet.model.models.unet_batchnorm(input_shape: object, loss: object, metrics: object, learning_rate: float = 0.0001, filter_size: float = 3, n_filters_factor: float = 1, n_forecast_days: int = 1, legacy_rounding: bool = False) → object

Parameters:

• input_shape

• loss

• metrics

• learning_rate

• filter_size

• n_filters_factor

• n_forecast_days

• legacy_rounding – Ensures filter number calculations are int()’d at the end of calculations

Returns:

icenet.model.predict module

icenet.model.predict.date_arg(string: str) → object

Parameters:

string

Returns:

icenet.model.predict.get_args()

Returns:

icenet.model.predict.main()

icenet.model.predict.predict_forecast(dataset_config: object, network_name: object, dataset_name: object = None, legacy_rounding: bool = False, model_func: callable = <function unet_batchnorm>, n_filters_factor: float = 0.125, network_folder: object = None, output_folder: object = None, save_args: bool = False, seed: int = 42, start_dates: object = (datetime.date(2025, 2, 3), ), test_set: bool = False) → object

Parameters:

• dataset_config

• network_name

• dataset_name

• legacy_rounding

• model_func

• n_filters_factor

• network_folder

• output_folder

• save_args

• seed

• start_dates

• test_set

Returns:

icenet.model.predict.run_prediction(network, date, output_folder, data_sample, save_args)

icenet.model.train module

icenet.model.train.evaluate_model(model_path: object, dataset: object, dataset_ratio: float = 1.0, max_queue_size: int = 3, workers: int = 5, use_multiprocessing: bool = True)

Parameters:

• model_path

• dataset

• dataset_ratio

• max_queue_size

• workers

• use_multiprocessing

icenet.model.train.get_args()

Returns:

icenet.model.train.main()

icenet.model.train.train_model(run_name: object, dataset: object, callback_objects: list = [], checkpoint_monitor: str = 'val_rmse', checkpoint_mode: str = 'min', dataset_ratio: float = 1.0, early_stopping_patience: int = 30, epochs: int = 2, filter_size: float = 3, learning_rate: float = 0.0001, lr_10e_decay_fac: float = 1.0, lr_decay_start: float = 10, lr_decay_end: float = 30, max_queue_size: int = 3, model_func: object = <function unet_batchnorm>, n_filters_factor: float = 2, network_folder: object = None, network_save: bool = True, pickup_weights: bool = False, pre_load_network: bool = False, pre_load_path: object = None, seed: int = 42, strategy: object = <tensorflow.python.distribute.distribute_lib._DefaultDistributionStrategy object>, training_verbosity: int = 1, workers: int = 5, use_multiprocessing: bool = True, use_tensorboard: bool = True) → object

Parameters:

• run_name

• dataset

• callback_objects

• checkpoint_monitor

• checkpoint_mode

• dataset_ratio

• early_stopping_patience

• epochs

• filter_size

• learning_rate

• lr_10e_decay_fac

• lr_decay_start

• lr_decay_end

• max_queue_size

• model_func

• n_filters_factor

• network_folder

• network_save

• pickup_weights

• pre_load_network

• pre_load_path

• seed

• strategy

• training_verbosity

• workers

• use_multiprocessing

• use_tensorboard

Returns:

icenet.model.utils module

icenet.model.utils.arr_to_ice_edge_arr(arr: object, thresh: object, land_mask: object, region_mask: object) → object

Compute a boolean mask with True over ice edge contour grid cells using matplotlib.pyplot.contour and an input threshold to define the ice edge (e.g. 0.15 for the 15% SIC ice edge or 0.5 for SIP forecasts). The contour along the coastline is removed using the region mask. :param arr: :param thresh: :param land_mask: :param region_mask: :return:

icenet.model.utils.arr_to_ice_edge_rgba_arr(arr: object, thresh: object, land_mask: object, region_mask: object, rgb: object) → object

Parameters:

• arr

• thresh

• land_mask

• region_mask

• rgb

Returns:

icenet.model.utils.compute_heatmap(results_df: object, model: object, seed: object = 'NA', metric: object = 'Binary accuracy') → object

Returns a binary accuracy heatmap of lead time vs. calendar month for a given model. :param results_df: :param model: :param seed: :param metric: :return:

icenet.model.utils.make_exp_decay_lr_schedule(rate: float, start_epoch: int = 1, end_epoch: object = inf) → object

Returns an exponential learning rate function that multiplies by exp(-rate) each epoch after start_epoch. :param rate: :param start_epoch: :param end_epoch: :return:

Module contents