A Keras Callback that uploads your Tensorboard logs to a Cloud Bucket
Currently, only Google Cloud Platform Storage is supported. Very little effort is needed to support AWS S3, so feel free to contribute to this project.
pip install keras-bucket-tensorboard-callback
The following example trains uploads the Tensorboard logs to you GCP Storage
bucket my-bucket
, inside the directory any_dir
:
# Import the class
from keras_bucket_tensorboard_callback import BucketTensorBoard
# Create the callback instance, passing the bucket URI
bucket_callback = BucketTensorBoard('gs://my-bucket/any_dir')
# Train the model with the callback
model.fit(
x=X,
y=Y,
epochs=20,
callbacks=[bucket_callback]
)
Make sure you have access to the provided bucket. For GCP, you should have
the GOOGLE_APPLICATION_CREDENTIALS
env set, pointing to your json
key
file.
With tensorboard installed your environment, run:
tensorboard --logdir=gs://my-bucket/any_dir
The TensorBoard will show your metrics and graphs saved on the bucket.