Executing batch experiments in Watson Studio was designed to be flexibile and to fit into a range of pre-existing workflows. However this flexibility can be daunting for new users. Now, executing experiments is as simple as:
experiment = Experiment("My experiment", "Test hyperparameter range",
"tensorflow", "1.5", "python","3.5",
studio_utils, project_utils)
experiment.add_training_run("cnn tests", hyperparams, "python3 experiment.py", "experiment.zip", "v100x2")
experiment.execute()
This repository's utility classes (below) hide the underlying complexity while providing the source code so you can dig deeper as you advance.
To help you get started, sample experiments are available for standard use cases. These samples are designed for quick execution from the command prompt but can be easily inserted into a notebook if desired.
The next section provides detailed setup instructions to complete these steps. (a) clone this repository then (b) install the Watson Machine Learning (WML) python client. (c) copy credentials to wml_credentials.json and cos_credentials.json then (d) execute the sample batch experiments.
If you are new to Watson Studio or simply want more details on confguring the credentials files, then follow these steps:
- Clone this repository using the "Clone or Download" button above
- Setup your IBM Cloud developer utilities
- Create Watson Machine Learning (WML) services + credentials
- Option A: Using Watson Studio's UI
- Option B: Using IBM Cloud CLI
- Install WML's tooling
- Create Cloud Object Storage (COS) service + credentials
- Option A: Using Watson Studio's UI
- Option B: Using IBM Cloud CLI
- Install Python client for COS
- Create a project in Watson Studio and save the project id
- Execute example batch experiments