tfwatcher.callbacks package

Submodules

tfwatcher.callbacks.epoch module

class tfwatcher.callbacks.epoch.EpochEnd(schedule: Union[int, list] = 1, round_time: int = 2, print_logs: bool = False)[source]

Bases: keras.callbacks.Callback

This class is a subclass of the tf.keras.callbacks.Callback abstract base class and overrides the methods on_epoch_begin() and on_epoch_end() allowing logging after epochs in training. This class also uses the firebase_helpers to send data to Firebase Realtime database and also creates a 7 character unique string where the data is pushed on Firebase. Logging to Firebase is also controllable by schedule argument, even providing a granular control for each epoch.

Example:

Logging data after every epoch
 1 import tfwatcher
 2 
 3 # here we specify schedule = 1 to log after every epoch
 4 monitor_callback = tfwatcher.callbacks.EpochEnd(schedule=1)
 5 
 6 model.compile(
 7     optimizer=...,
 8     loss=...,
 9     # metrics which will be logged
10     metrics=[...],
11 )
12 
13 model.fit(..., callbacks=[monitor_callback])
Parameters

  • schedule (Union[int, list[int]], optional) – Use an integer value n to specify logging data every n epochs the first one being logged by default. Use a list of integers to control logging with a greater granularity, logs on all epoch numbers specified in the list taking the first epoch as epoch 1. Using a list will override loggging on the first epoch by default, defaults to 1

  • round_time (int, optional) – This argument allows specifying if you want to see the times on the web-app to be rounded, in most cases you would not be using this, defaults to 2

  • print_logs (bool, optional) – This argument should only be used when trying to debug if your logs do not appear in the web-app, if set to True this would print out the dictionary which is being pushed to Firebase, defaults to False

Raises

  • ValueError – If the schedule is neither an integer or a list.

  • Exception – If all the values in schedule list are not convertible to integer.

on_epoch_begin(epoch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_epoch_begin method which is called at the start of an epoch. This function should only be called during TRAIN mode.

Parameters

  • epoch (int) – Index of epoch

  • logs (dict, optional) – Currently no data is passed to this argument since there are no logs during the start of an epoch, defaults to None

on_epoch_end(epoch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_epoch_end method which is called at the end of an epoch. This function should only be called during TRAIN mode. This method adds the epoch number, the average time taken and pushes it to Firebase using the firebase_helpers module.

Parameters

  • epoch (int) – Index of epoch

  • logs (dict, optional) – Metric results for this training epoch, and for the validation epoch if validation is performed. Validation result keys are prefixed with val_. For training epoch, the values of the Model’s metrics are returned. Example : {'loss': 0.2, 'accuracy': 0.7}, defaults to None

tfwatcher.callbacks.predict module

class tfwatcher.callbacks.predict.PredictEnd(round_time: int = 2, print_logs: bool = False)[source]

Bases: keras.callbacks.Callback

This class is a subclass of the tf.keras.callbacks.Callback abstract base class and overrides the methods on_predict_begin() and on_predict_end() allowing loging after predict method is run. This class also uses the firebase_helpers module to send data to Firebase Realtime database and also creates a 7 character unique string where the data is pushed on Firebase.

Note

This class does not have the schedule parameter like other clases in the tfwatcher.callbacks subpackage since this would notify you once the prediction is over and there are no batches or epochs to make a schedule for.

Example:

Logging data after predict method
 1 import tfwatcher
 2 
 3 monitor_callback = tfwatcher.callbacks.PredictEnd()
 4 
 5 model.compile(
 6     optimizer=...,
 7     loss=...,
 8     # metrics which will be logged
 9     metrics=[...],
10 )
11 
12 model.fit(..., callbacks=[monitor_callback])
Parameters

  • round_time (int, optional) – This argument allows specifying if you want to see the times on the web-app to be rounded, in most cases you would not be using this, defaults to 2

  • print_logs (bool, optional) – This argument should only be used when trying to debug if your logs do not appear in the web-app, if set to True this would print out the dictionary which is being pushed to Firebase, defaults to False

Raises

  • ValueError – If the schedule is neither an integer or a list.

  • Exception – If all the values in schedule list are not convertible to integer.

on_predict_begin(logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_predict_begin method which is called at the start of prediction.

Parameters

logs (dict, optional) – Currently no data is passed to this argument since there are no logs during the start of an epoch, defaults to None

on_predict_end(logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_predict_end method which is called at the end of prediction.

Parameters

logs (dict, optional) – Currently no data is passed to this argument since there are no logs during the start of an epoch, defaults to None

tfwatcher.callbacks.predict_batch module

class tfwatcher.callbacks.predict_batch.PredictBatchEnd(schedule: Union[int, list] = 1, round_time: int = 2, print_logs: bool = False)[source]

Bases: keras.callbacks.Callback

This class is a subclass of the tf.keras.callbacks.Callback abstract base class and overrides the methods on_predict_batch_begin() and on_predict_batch_end() allowing loging after batches in predict method. This class also uses the firebase_helpers to send data to Firebase Realtime database and also creates a 7 character unique string where the data is pushed on Firebase. Logging to Firebase is also controllable by schedule argument, even providing a granular control for each batch in predict methods.

Example:

Logging data after every batch in predict methods
 1 import tfwatcher
 2 
 3 # here we specify schedule = 1 to log after every batch
 4 monitor_callback = tfwatcher.callbacks.PredictBatchEnd(schedule=1)
 5 
 6 model.compile(
 7     optimizer=...,
 8     loss=...,
 9     # metrics which will be logged
10     metrics=[...],
11 )
12 
13 model.fit(..., callbacks=[monitor_callback])

Warning

If the steps_per_execution argument to compile in tf.keras.Model is set to N, the logging code will only be called every N batches.

Parameters

  • schedule (Union[int, list[int]], optional) – Use an integer value n to specify logging data every n batches the first one being logged by default. Use a list of integers to control logging with a greater granularity, logs on all batch numbers specified in the list taking the first batch as batch 1. Using a list will override loggging on the first batch by default, defaults to 1

  • round_time (int, optional) – This argument allows specifying if you want to see the times on the web-app to be rounded, in most cases you would not be using this, defaults to 2

  • print_logs (bool, optional) – This argument should only be used when trying to debug if your logs do not appear in the web-app, if set to True this would print out the dictionary which is being pushed to Firebase, defaults to False

Raises

  • ValueError – If the schedule is neither an integer or a list.

  • Exception – If all the values in schedule list are not convertible to integer.

on_predict_batch_begin(batch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_predict_batch_begin method which is called called at the beginning of a batch in predict methods.

Parameters

  • batch (int) – Index of batch within the current epoch

  • logs (dict, optional) – contains the return value of model.predict_step, it typically returns a dict with a key ‘outputs’ containing the model’s outputs

on_predict_batch_end(batch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_predict_batch_end method which is called called at the end of a batch in predict methods. This method adds the batch number, the average time taken and pushes it to Firebase using the firebase_helpers module.

Parameters

  • epoch (int) – Index of batch within the current epoch

  • logs (dict, optional) – Aggregated metric results up until this batch, defaults to None

tfwatcher.callbacks.test_batch module

class tfwatcher.callbacks.test_batch.TestBatchEnd(schedule: Union[int, list] = 1, round_time: int = 2, print_logs: bool = False)[source]

Bases: keras.callbacks.Callback

This class is a subclass of the tf.keras.callbacks.Callback abstract base class and overrides the methods on_test_batch_begin() and on_test_batch_end() allowing loging after batches in evaluate methods and at the beginning of a validation batch in the fit methods, if validation data is provided. This class also uses the firebase_helpers to send data to Firebase Realtime database and also creates a 7 character unique string where the data is pushed on Firebase. Logging to Firebase is also controllable by schedule argument, even providing a granular control for each batch in evaluate methods.

Example:

Logging data after every batch in evaluate methods
 1 import tfwatcher
 2 
 3 # here we specify schedule = 1 to log after every batch
 4 monitor_callback = tfwatcher.callbacks.TestBatchEnd(schedule=1)
 5 
 6 model.compile(
 7     optimizer=...,
 8     loss=...,
 9     # metrics which will be logged
10     metrics=[...],
11 )
12 
13 model.fit(..., callbacks=[monitor_callback])

Warning

If the steps_per_execution argument to compile in tf.keras.Model is set to N, the logging code will only be called every N batches.

Parameters

  • schedule (Union[int, list[int]], optional) – Use an integer value n to specify logging data every n batches the first one being logged by default. Use a list of integers to control logging with a greater granularity, logs on all batch numbers specified in the list taking the first batch as batch 1. Using a list will override loggging on the first batch by default, defaults to 1

  • round_time (int, optional) – This argument allows specifying if you want to see the times on the web-app to be rounded, in most cases you would not be using this, defaults to 2

  • print_logs (bool, optional) – This argument should only be used when trying to debug if your logs do not appear in the web-app, if set to True this would print out the dictionary which is being pushed to Firebase, defaults to False

Raises

  • ValueError – If the schedule is neither an integer or a list.

  • Exception – If all the values in schedule list are not convertible to integer.

on_test_batch_begin(batch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_test_batch_begin method which is called called at the beginning of a batch in evaluate methods and at the beginning of a validation batch in the fit methods, if validation data is provided.

Parameters

  • batch (int) – Index of batch within the current epoch

  • logs (dict, optional) – contains the return value of model.test_step. Typically, the values of the Model’s metrics are returned. Example: {'loss': 0.2, 'accuracy': 0.7}.

on_test_batch_end(batch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_test_batch_end method which is called called at the end of a batch in evaluate methods and at the beginning of a validation batch in the fit methods, if validation data is provided. This method adds the batch number, the average time taken and pushes it to Firebase using the firebase_helpers module.

Parameters

  • epoch (int) – Index of batch within the current epoch

  • logs (dict, optional) – Aggregated metric results up until this batch, defaults to None

tfwatcher.callbacks.train_batch module

class tfwatcher.callbacks.train_batch.TrainBatchEnd(schedule: Union[int, list] = 1, round_time: int = 2, print_logs: bool = False)[source]

Bases: keras.callbacks.Callback

This class is a subclass of the tf.keras.callbacks.Callback abstract base class and overrides the methods on_train_batch_begin() and on_train_batch_end() allowing loging after a training batch in fit methods. This class also uses the firebase_helpers to send data to Firebase Realtime database and also creates a 7 character unique string where the data is pushed on Firebase. Logging to Firebase is also controllable by schedule argument, even providing a granular control for each batch in fit methods.

Example:

Logging data after every batch in fit methods
 1 import tfwatcher
 2 
 3 # here we specify schedule = 1 to log after every batch
 4 monitor_callback = tfwatcher.callbacks.TrainBatchEnd(schedule=1)
 5 
 6 model.compile(
 7     optimizer=...,
 8     loss=...,
 9     # metrics which will be logged
10     metrics=[...],
11 )
12 
13 model.fit(..., callbacks=[monitor_callback])

Warning

If the steps_per_execution argument to compile in tf.keras.Model is set to N, the logging code will only be called every N batches.

Parameters

  • schedule (Union[int, list[int]], optional) – Use an integer value n to specify logging data every n batches the first one being logged by default. Use a list of integers to control logging with a greater granularity, logs on all batch numbers specified in the list taking the first batch as batch 1. Using a list will override loggging on the first batch by default, defaults to 1

  • round_time (int, optional) – This argument allows specifying if you want to see the times on the web-app to be rounded, in most cases you would not be using this, defaults to 2

  • print_logs (bool, optional) – This argument should only be used when trying to debug if your logs do not appear in the web-app, if set to True this would print out the dictionary which is being pushed to Firebase, defaults to False

Raises

  • ValueError – If the schedule is neither an integer or a list.

  • Exception – If all the values in schedule list are not convertible to integer.

on_train_batch_begin(batch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_train_batch_begin method which is called called at the beginning of a training batch in fit methods.

Parameters

  • batch (int) – Index of batch within the current epoch

  • logs (dict, optional) – Contains the return value of model.train_step. Typically, the values of the Model’s metrics are returned. Example: {'loss': 0.2, 'accuracy': 0.7}, defaults to None

on_train_batch_end(batch: int, logs: Optional[dict] = None)[source]

Overrides the tf.keras.callbacks.Callback.on_train_batch_end method which is called called at the end of a batch in of a training batch in fit methods. This method adds the batch number, the average time taken and pushes it to Firebase using the firebase_helpers module.

Parameters

  • epoch (int) – Index of batch within the current epoch

  • logs (dict, optional) – Aggregated metric results up until this batch, defaults to None

Module contents