Keras

Try in a Colab Notebook here →

The Weights & Biases Keras Callbacks

We have added three new callbacks for Keras and TensorFlow users, available from wandb v0.13.4. For the legacy WandbCallback scroll down.

WandbMetricsLogger : Use this callback for Experiment Tracking. It will log your training and validation metrics along with system metrics to Weights and Biases.

WandbModelCheckpoint : Use this callback to log your model checkpoints to Weight and Biases Artifacts.

WandbEvalCallback: This base callback will log model predictions to Weights and Biases Tables for interactive visualization.

These new callbacks,

• Adhere to Keras design philosophy
• Reduce the cognitive load of using a single callback (WandbCallback) for everything,
• Make it easy for Keras users to modify the callback by subclassing it to support their niche use case.

Experiment Tracking with WandbMetricsLogger

Try in a Colab Notebook here →

WandbMetricsLogger automatically logs Keras' logs dictionary that callback methods such as on_epoch_end, on_batch_end etc, take as an argument.

Using this provides:

• train and validation metrics defined in model.compile
• system (CPU/GPU/TPU) metrics
• learning rate (both for a fixed value or a learning rate scheduler)

import wandb
from wandb.integration.keras import WandbMetricsLogger

# Initialize a new W&B run
wandb.init(config={"bs": 12})

# Pass the WandbMetricsLogger to model.fit
model.fit(
    X_train, y_train, validation_data=(X_test, y_test), callbacks=[WandbMetricsLogger()]
)

WandbMetricsLogger Reference

Parameter	Description
log_freq	("epoch", "batch", or int): if "epoch", logs metrics at the end of each epoch. If "batch", logs metrics at the end of each batch. If an int, logs metrics at the end of that many batches. Defaults to "epoch".
initial_global_step	(int): Use this argument to correctly log the learning rate when you resume training from some initial_epoch, and a learning rate scheduler is used. This can be computed as step_size * initial_step. Defaults to 0.

Model Checkpointing using WandbModelCheckpoint

Try in a Colab Notebook here →

Use WandbModelCheckpoint callback to save the Keras model (SavedModel format) or model weights periodically and uploads them to W&B as a wandb.Artifact for model versioning.

This callback is subclassed from tf.keras.callbacks.ModelCheckpoint ,thus the checkpointing logic is taken care of by the parent callback.

This callback provides the following features:

• Save the model that has achieved "best performance" based on the "monitor".
• Save the model at the end of every epoch regardless of the performance.
• Save the model at the end of the epoch or after a fixed number of training batches.
• Save only model weights, or save the whole model.
• Save the model either in SavedModel format or in .h5 format.

This callback should be used in conjunction with WandbMetricsLogger.

import wandb
from wandb.integration.keras import WandbMetricsLogger, WandbModelCheckpoint

# Initialize a new W&B run
wandb.init(config={"bs": 12})

# Pass the WandbModelCheckpoint to model.fit
model.fit(
    X_train,
    y_train,
    validation_data=(X_test, y_test),
    callbacks=[
        WandbMetricsLogger(),
        WandbModelCheckpoint("models"),
    ],
)

WandbModelCheckpoint Reference

Parameter	Description
filepath	(str): path to save the mode file.
monitor	(str): The metric name to monitor.
verbose	(int): Verbosity mode, 0 or 1. Mode 0 is silent, and mode 1 displays messages when the callback takes an action.
save_best_only	(bool): if save_best_only=True, it only saves when the model is considered the "best" and the latest best model according to the quantity monitored (monitor) will not be overwritten.
save_weights_only	(bool): if True, then only the model's weights will be saved.
mode	("auto", "min", or "max"): For val_acc, this should be ‘max’, for val_loss this should be ‘min’, etc.
save_weights_only	(bool): if True, then only the model's weights will be saved.
save_freq	("epoch" or int): When using ‘epoch’, the callback saves the model after each epoch. When using an integer, the callback saves the model at end of this many batches. Note that when monitoring validation metrics such as val_acc or val_loss, save_freq must be set to "epoch" as those metrics are only available at the end of an epoch.
options	(str): Optional tf.train.CheckpointOptions object if save_weights_only is true or optional tf.saved_model.SaveOptions object if save_weights_only is false.
initial_value_threshold	(float): Floating point initial "best" value of the metric to be monitored.

How to log checkpoints after N epochs?

By default (save_freq="epoch") the callback creates a checkpoint and uploads it as an artifact after each epoch. If we pass an integer to save_freq the checkpoint will be created after that many batches. To checkpoint after N epochs, compute the cardinality of the train dataloader and pass it to save_freq:

WandbModelCheckpoint(
    filepath="models/",
    save_freq=int((trainloader.cardinality()*N).numpy())
)

How to log checkpoints on a TPU Node architecture efficiently?

While checkpointing on TPUs you might encounter UnimplementedError: File system scheme '[local]' not implemented error message. This happens because the model directory (filepath) must use a cloud storage bucket path (gs://bucket-name/...), and this bucket must be accessible from the TPU server. We can however, use the local path for checkpointing which in turn is uploaded as an Artifacts.

checkpoint_options = tf.saved_model.SaveOptions(experimental_io_device="/job:localhost")

WandbModelCheckpoint(
    filepath="models/,
    options=checkpoint_options,
)

Model Prediction Visualization using WandbEvalCallback

Try in a Colab Notebook here →

The WandbEvalCallback is an abstract base class to build Keras callbacks primarily for model prediction and, secondarily, dataset visualization.

This abstract callback is agnostic with respect to the dataset and the task. To use this, inherit from this base WandbEvalCallback callback class and implement the add_ground_truth and add_model_prediction methods.

The WandbEvalCallback is a utility class that provides helpful methods to:

• create data and prediction wandb.Table instances,
• log data and prediction Tables as wandb.Artifact
• logs the data table on_train_begin
• logs the prediction table on_epoch_end

For example, we have implemented WandbClfEvalCallback below for an image classification task. This example callback:

• logs the validation data (data_table) to W&B,
• performs inference and logs the prediction (pred_table) to W&B at the end of every epoch.

import wandb
from wandb.integration.keras import WandbMetricsLogger, WandbEvalCallback


# Implement your model prediction visualization callback
class WandbClfEvalCallback(WandbEvalCallback):
    def __init__(
        self, validation_data, data_table_columns, pred_table_columns, num_samples=100
    ):
        super().__init__(data_table_columns, pred_table_columns)

        self.x = validation_data[0]
        self.y = validation_data[1]

    def add_ground_truth(self, logs=None):
        for idx, (image, label) in enumerate(zip(self.x, self.y)):
            self.data_table.add_data(idx, wandb.Image(image), label)

    def add_model_predictions(self, epoch, logs=None):
        preds = self.model.predict(self.x, verbose=0)
        preds = tf.argmax(preds, axis=-1)

        table_idxs = self.data_table_ref.get_index()

        for idx in table_idxs:
            pred = preds[idx]
            self.pred_table.add_data(
                epoch,
                self.data_table_ref.data[idx][0],
                self.data_table_ref.data[idx][1],
                self.data_table_ref.data[idx][2],
                pred,
            )


# ...

# Initialize a new W&B run
wandb.init(config={"hyper": "parameter"})

# Add the Callbacks to Model.fit
model.fit(
    X_train,
    y_train,
    validation_data=(X_test, y_test),
    callbacks=[
        WandbMetricsLogger(),
        WandbClfEvalCallback(
            validation_data=(X_test, y_test),
            data_table_columns=["idx", "image", "label"],
            pred_table_columns=["epoch", "idx", "image", "label", "pred"],
        ),
    ],
)

info

💡 The Tables are logged to the W&B Artifact page by default and not the Workspace page.

WandbEvalCallback Reference

Parameter	Description
data_table_columns	(list) List of column names for the data_table
pred_table_columns	(list) List of column names for the pred_table

How the memory footprint is reduced?

We log the data_table to W&B when the on_train_begin method is invoked. Once it's uploaded as a W&B Artifact, we get a reference to this table which can be accessed using data_table_ref class variable. The data_table_ref is a 2D list that can be indexed like self.data_table_ref[idx][n], where idx is the row number while n is the column number. Let's see the usage in the example below.

Customize the callback further

You can override the on_train_begin or on_epoch_end methods to have more fine-grained control. If you want to log the samples after N batches, you can implement on_train_batch_end method.

info

💡 If you are implementing a callback for model prediction visualization by inheriting WandbEvalCallback and something needs to be clarified or fixed, please let us know by opening an issue.

WandbCallback [Legacy]

Use the W&B library WandbCallback Class to automatically save all the metrics and the loss values tracked in model.fit.

import wandb
from wandb.integration.keras import WandbCallback

wandb.init(config={"hyper": "parameter"})

...  # code to set up your model in Keras

# Pass the callback to model.fit
model.fit(
    X_train, y_train, validation_data=(X_test, y_test), callbacks=[WandbCallback()]
)

Usage examples

See this one minute, step-by-step video if this is your first time integrating W&B with Keras: Get Started with Keras and Weights & Biases in Less Than a Minute

For a more detailed video, see Integrate Weights & Biases with Keras. The notebook example used can be found here: Colab Jupyter Notebook.

info

Try W&B and Keras integration example from the video above in a colab notebook. Or see our example repo for scripts, including a Fashion MNIST example and the W&B Dashboard it generates.

The WandbCallback class supports a wide variety of logging configuration options: specifying a metric to monitor, tracking of weights and gradients, logging of predictions on training_data and validation_data, and more.

Check out the reference documentation for the keras.WandbCallback for full details.

The WandbCallback

• will automatically log history data from any metrics collected by keras: loss and anything passed into keras_model.compile()
• will set summary metrics for the run associated with the "best" training step, where "best" is defined by the monitor and mode attributes. This defaults to the epoch with the minimum val_loss. WandbCallback will by default save the model associated with the best epoch
• can optionally log gradient and parameter histogram
• can optionally save training and validation data for wandb to visualize.

WandbCallback Reference

Arguments
monitor	(str) name of metric to monitor. Defaults to val_loss.
mode	(str) one of {auto, min, max}. min - save model when monitor is minimized max - save model when monitor is maximized auto - try to guess when to save the model (default).
save_model	True - save a model when monitor beats all previous epochs False - don't save models
save_graph	(boolean) if True save model graph to wandb (default to True).
save_weights_only	(boolean) if True, then only the model's weights will be saved (model.save_weights(filepath)), else the full model is saved (model.save(filepath)).
log_weights	(boolean) if True save histograms of the model's layer's weights.
log_gradients	(boolean) if True log histograms of the training gradients
training_data	(tuple) Same format (X,y) as passed to model.fit. This is needed for calculating gradients - this is mandatory if log_gradients is True.
validation_data	(tuple) Same format (X,y) as passed to model.fit. A set of data for wandb to visualize. If this is set, every epoch, wandb will make a small number of predictions and save the results for later visualization.
generator	(generator) a generator that returns validation data for wandb to visualize. This generator should return tuples (X,y). Either validate_data or generator should be set for wandb to visualize specific data examples.
validation_steps	(int) if validation_data is a generator, how many steps to run the generator for the full validation set.
labels	(list) If you are visualizing your data with wandb this list of labels will convert numeric output to understandable string if you are building a multiclass classifier. If you are making a binary classifier you can pass in a list of two labels ["label for false", "label for true"]. If validate_data and generator are both false, this won't do anything.
predictions	(int) the number of predictions to make for visualization each epoch, max is 100.
input_type	(string) type of the model input to help visualization. can be one of: (image, images, segmentation_mask).
output_type	(string) type of the model output to help visualziation. can be one of: (image, images, segmentation_mask).
log_evaluation	(boolean) if True, save a Table containing validation data and the model's predictions at each epoch. See validation_indexes, validation_row_processor, and output_row_processor for additional details.
class_colors	([float, float, float]) if the input or output is a segmentation mask, an array containing an rgb tuple (range 0-1) for each class.
log_batch_frequency	(integer) if None, callback will log every epoch. If set to integer, callback will log training metrics every log_batch_frequency batches.
log_best_prefix	(string) if None, no extra summary metrics will be saved. If set to a string, the monitored metric and epoch will be prepended with this value and stored as summary metrics.
validation_indexes	([wandb.data_types._TableLinkMixin]) an ordered list of index keys to associate with each validation example. If log_evaluation is True and validation_indexes is provided, then a Table of validation data will not be created and instead each prediction will be associated with the row represented by the TableLinkMixin. The most common way to obtain such keys are is use Table.get_index() which will return a list of row keys.
validation_row_processor	(Callable) a function to apply to the validation data, commonly used to visualize the data. The function will receive an ndx (int) and a row (dict). If your model has a single input, then row["input"] will be the input data for the row. Else, it will be keyed based on the name of the input slot. If your fit function takes a single target, then row["target"] will be the target data for the row. Else, it will be keyed based on the name of the output slots. For example, if your input data is a single ndarray, but you wish to visualize the data as an Image, then you can provide lambda ndx, row: {"img": wandb.Image(row["input"])} as the processor. Ignored if log_evaluation is False or validation_indexes are present.
output_row_processor	(Callable) same as validation_row_processor, but applied to the model's output. row["output"] will contain the results of the model output.
infer_missing_processors	(bool) Determines if validation_row_processor and output_row_processor should be inferred if missing. Defaults to True. If labels are provided, we will attempt to infer classification-type processors where appropriate.
log_evaluation_frequency	(int) Determines the frequency which evaluation results will be logged. Default 0 (only at the end of training). Set to 1 to log every epoch, 2 to log every other epoch, and so on. Has no effect when log_evaluation is False.

Frequently Asked Questions

How do I use Keras multiprocessing with wandb?

If you're setting use_multiprocessing=True and seeing an error like:

Error("You must call wandb.init() before wandb.config.batch_size")

then try this:

1. In the Sequence class construction, add: wandb.init(group='...')
2. In your main program, make sure you're using if __name__ == "__main__": and then put the rest of your script logic inside that.