Improving my deep learning testing syste

Improving my deep learning testing system In cs224n I built a test rig for tuning hyperparameters on deep learning models. I want to extend this for this quarter. 

1. There should be a server/client architecture, where tests are submitted centrally and farmed out to devices. Must include universal identifier for each test, and provide more context for test results. 

2. I want to specify the dimensions on which hyperparameters may be tuned. A model should be able to specify a hyperparameter’s range, including: 

3. Sampling distribution (categorical, ordinal, logarithmic…)

4. Bounds

5. Estimated correlation with other factors. (this determines order of testing)