Training#

You can provide data in various way to train your model: as a numpy array, using a path to a folder or files, or by using CAREamics data module class for more control (advanced).

The details of how CAREamics deals with the loading and patching is detailed in the dataset section.

Training by passing an array#

CAREamics can be trained by simply passing numpy arrays.

import numpy as np

train_array = np.random.rand(256, 256)
val_array = np.random.rand(256, 256)

careamist.train(
    train_source=train_array,  # (1)!
    val_source=val_array,  # (2)!
)

1. All parameters to the train method must be specified by keyword.
2. If you don't provide a validation source, CAREamics will use a fraction of the training data to validate the model.

careamist_supervised.train(
    train_source=train_array,
    train_target=target_array,
    val_source=val_array,
    val_target=val_target_array,
)

Training by passing a path#

The same thing can be done by passing a path to a folder or files.

careamist.train(
    train_source=path_to_train_data,  # (1)!
    val_source=path_to_val_data,
)

1. The path can point to a single file, or contain multiple files.

Splitting validation from training data#

If you only provide training data, CAREamics will extract the validation data directly from the training set. There are two parameters controlling that behaviour: val_percentage and val_minimum_split.

val_percentage is the fraction of the training data that will be used for validation, and val_minimum_split is the minimum number of images used. If the percentage leads to a number of patches smaller than val_minimum_split, CAREamics will use val_minimum_split.

careamist.train(
    train_source=train_array,
    val_percentage=0.1,  # (1)!
    val_minimum_split=5,  # (2)!
)

1. 10% of the training data will be used for validation.
2. If the number of images is less than 5, CAREamics will use 5 images for validation.

Training by passing a TrainDataModule object#

CAREamics provides a class to handle the data loading of custom data type. We will dive in more details in the next section into what this class can be used for. Here is a brief overview of how it is passed to the train method.

from careamics.lightning import TrainDataModule

data_module = TrainDataModule(  # (1)!
    data_config=config.data_config, train_data=train_array
)

careamist.train(datamodule=data_module)

1. Here this does the same thing as passing the train_source directly into the train method. In the next section, we will see a more useful example.

Logging the training#

By default, CAREamics simply log the training progress in the console. However, it is possible to use either WandB or TensorBoard.

To decide on the logger, check out the Configuration section.