some changes of the hit rate plot?

[LilySnow]

some changes of the hit rate plot?

[NicoRenaud]

1. I modified the code so that the hit plot data can be exported at each epoch in the hdf5 output file (still have to push it). I've modified DeepXplorer so that we can plot the hit rate either for individual epoch or multiple epoch on one graph.

2. That's what's done already no ?

3. We'll have to somehow include the HADDOCK score and corresponding hit plot to the output hdf5 file Another solution would be to have an external script that reads the HDF5 file and the Haddock score and plot all that.

[LilySnow]

1. Thanks. I have not checked the DeepXplorer yet.
2. Sorry, I have not pulled yet ... Will check it out.
3. Maybe include haddock score to the input hdf5 files? But you are the expert.

[NicoRenaud]

there is now a new option to save all the epoch in the output hdf5 file

model.train(....,save_epoch='all')

Then in the .hdf5 file there will be one group for each epoch. In this group there are the train/valid/test subgroup and each subgroup will contain the hit rate data.

You can vizualize all that with the DeepXplorer. You can select mulitple epoch group and right click. You'll have the option to plot the hit plot for the seleted epochs. I've also uploaded a hdf5 file containing the haddock score. If you also select this one with the epoch it will also plot the hit rate of the haddock score.

[LilySnow]

Just pull the master branch. The hit rate plot still plots top N, where N is the training data, instead of the test data.

[NicoRenaud]

But in that case we will only see a portion of the trainging and validatiton plot no ?
We could have either:
a percentage of the x axis so that training/validatation/test have the same x axis
individual plot for the training/validatation/test cases
What do you think ?