New tool to run batch jobs, test hyperparameters, monitor training remotely

[Nate Harada]

Hi Everyone,

I've been using Caffe for my research, but have always had trouble finding a good way to test hyperparameters, as well as comparing the results of those runs. 

I'd like to introduce Mri, which is a tool I just open-sourced that automatically parses Caffe output and sends results to a web-based dashboard accessible from anywhere. Included is a script to automatically generate new model and solver files based on hyperparameter search. Check it out here: http://mri.readthedocs.org/en/latest/

Feel free to ask me any questions, and especially to submit bug reports. I'm happy to help out!

Best,

Nate Harada

[Patrick Grossmann]

Interesting! I'm excited to test this tool, would be a very useful extension. Question, does PyCaffe allow writing solver.txt files specifying the learning parameters or only the prototxt specifying the architecture?

[Nate Harada]

does PyCaffe allow writing solver.txt files specifying the learning parameters

Not as far as I know, although since the solver files are much simpler to parse and create rolling your own probably wouldn't be too bad. It would be nice though if there was a Python function that was aware of the possible values for the solver.