- Changed message protocol to MsgPack
- Made Keras 2 compatible
- Fully Python3 compatibility
- Fixed issue with generating image insights for Keras 1 in Tensorflow
- Added new file upload method
- Added new batch progress method
- Many bug fixes and improvements
See http://aetros.com/docu/python-sdk/api/aetros.backend.jobbackend for full API documentation
Changed Keras integration API
Please use now our "Keras Callback"
python
import aetros.backend
job = aetros.backend.start_job('name/model')
(X_train, y_train), (X_test, y_test) = mnist.load_data()
model = Sequential()
model.add(...)
model.compile(...)
version simplest
aetros_callback = job.create_keras_callback(model)
or version with insights images and confusion matrix
aetros_callback = job.create_keras_callback(
model,
insights=True, insights_x=X_train[0],
confusion_matrix=True, validation_data=(X_test, Y_test)
)
model.fit(
X_train, Y_train, batch_size=16, nb_epoch=50,
validation_data=(X_test, Y_test), callbacks=[aetros_callback]
)