Once the model file is set up, we need to provide the configuration file in the same trainer folder so that ML engine knows the parameters that need to be tuned, as well as the typical min and max values of the parameter.
We create the configuration file as follows in the trainer folder:
vim hptune.yaml
The following code is inserted into the preceding file:
trainingInput: pythonVersion: "3.5" scaleTier: CUSTOM masterType: standard_gpu hyperparameters: goal: MAXIMIZE hyperparameterMetricTag: accuracy maxTrials: 10 maxParallelTrials: 1 params: - parameterName: num-steps type: INTEGER minValue: 200 maxValue: 10000 scaleType: UNIT_LINEAR_SCALE - parameterName: lr type: DOUBLE minValue: 0.001 maxValue: 0.1 scaleType: UNIT_LOG_SCALE ...
