Hi! We’ve created a small command line utility that will allow users to illustrate the current state of a network/model using mxnet by graphing its validation and training data from a log file. The program will create two graphs showcasing the accuracy and loss of each epoch, allowing for data analysis that should be much simpler to interpret. We hope to see this tool be useful to anyone looking to better visualize their network data. Enjoy!
Useful visualizations @alexbenasutti!
I usually write my own logs in a custom format. Is there an MXNet standard log format you’re using?
Also have you seen the MXBoard project? https://github.com/awslabs/mxboard
And there’s a good blog post here.
Thanks @thomelane !
The log format is the standard log generated by the Module API for the fit method.
Also, the MXBoard project looks very interesting! While I currently don’t use TensorBoard, it’s definitely something I’ll look into.