Skip to content

Keras wrapper that autosaves what ModelCheckpoint cannot.

License

Notifications You must be signed in to change notification settings

dorukkarinca/keras-buoy

Repository files navigation

keras-buoy

https://travis-ci.com/dorukkarinca/keras-buoy.svg?branch=master

Keras wrapper that autosaves and auto-recovers not just the model weights but also the last epoch number and training history metrics.

See it in action in this Colab notebook!

pip install keras-buoy

Description

When training is interrupted and you rerun the whole code, it recovers the model weights and the epoch counter to the last saved values. Then it resumes training as if nothing happened. At the end, the Keras History.history dictionaries are combined so that the training history looks like one single training run.

Example

>>> from tensorflow import keras
>>> from keras_buoy.models import ResumableModel

>>> model = keras.Sequential()
...
>>> resumable_model = ResumableModel(model,
                                     save_every_epochs=4,
                                     custom_objects=None,
                                     to_path='/path/to/save/model_weights.h5')
>>> history = resumable_model.fit(x=x_train,
                                  y=y_train,
                                  validation_split=0.1,
                                  batch_size=256,
                                  verbose=2,
                                  epochs=15)

Recovered model from kerascheckpoint.h5 at epoch 8.

Epoch 9/15
1125/1125 - 5s - loss: 0.4790 - top_k_categorical_accuracy: 0.9698 - val_loss: 1.1075 - val_top_k_categorical_accuracy: 0.9206
Epoch 10/15
1125/1125 - 5s - loss: 0.4758 - top_k_categorical_accuracy: 0.9701 - val_loss: 1.1119 - val_top_k_categorical_accuracy: 0.9214
Epoch 11/15
1125/1125 - 5s - loss: 0.4753 - top_k_categorical_accuracy: 0.9702 - val_loss: 1.1000 - val_top_k_categorical_accuracy: 0.9215
Epoch 12/15

Try it out yourself in this Colab notebook.

Docs

keras_buoy.models.ResumableModel

Creates a resumable model.

Parameters:

Parameter name Description
model (tf.keras.Model) The instance of tf.keras.Model which you want to make resumable.
save_every_epochs (int) Specifies how often to save the model, history, and epoch counter. In case of a crash, recovery will happen from the last saved epoch multiple.
custom_objects (dict) At recovery time, this is passed into tf.keras.models.load_model(...) exactly as shown in Tensorflow docs so you can load your model with a custom loss for example.
to_path (str) Specifies the path where the model weights will be saved. If it ends with .h5, then it saves in the Keras H5 format instead of the default TensorFlow SavedModel format.

If to_path is mymodel.h5, then there will be mymodel_epoch_num.pkl and mymodel_history.pkl in the same directory as mymodel.h5, which hold backups for the epoch counter and the history dict, respectively.

Returns:

A ResumableModel instance. You can call .fit(...) on it.




keras.buoy.models.ResumableModel.fit

Fits a resumable model.

Parameters:

The accepted parameters are the same as tf.Keras.model.fit(...) except you cannot specify initial_epoch.

Returns:

history (dict): The history dict of the Keras History object. Note that it does not return the Keras.History object itself, just the dict.

Note

This project has been set up using PyScaffold 3.2.3. For details and usage information on PyScaffold see https://pyscaffold.org/.