Release 0.2.2
Features
- this release makes Maggy ready for Hopsworks 0.10.0
- Adds a SingleRun optimizer so users can run model training only once with
experiment.lagom(train)
- It is now possible to run multiple Maggy experiments from the same yarn app with the progress information and logging
- Using
print
in the training wrapper function will propagate the prints from the Spark Executors to Jupyter and display them underneath the Jupyter cell.The0: Train on 60000 samples, validate on 10000 samples 1: x_train shape: (60000, 28, 28, 1)
0:
and1:
indicate from which machine the prints are coming.
This feature should be used with care.
Bugfixes
- Fixes a serialization Error when user function returns numpy data type