update paper data example #245
Conversation
Pull Request Test Coverage Report for Build 1611058010
💛 - Coveralls |
deeprank/learn/NeuralNet.py
Outdated
| try: | ||
| self.hit_cutoff = self.state['hit_cutoff'] | ||
| except Exception: | ||
| print('No hit_cutoff provided') |
There was a problem hiding this comment.
Please use logger to output messages.
I guess this Exception only applies to our published models. If so, we have to tell users what to do in the message, like No "hit_cutoff" found in {model}. Please set it in function "test()" when doing prediction".
There was a problem hiding this comment.
The hit cutoff is used only if target values are provided. In prediction mode you do not need it, you need it for benchmark only.
deeprank/learn/NeuralNet.py
Outdated
| @@ -766,7 +770,8 @@ def _train(self, index_train, index_valid, index_test, | |||
| return torch.cat([param.data.view(-1) | |||
| for param in self.net.parameters()], 0) | |||
|
|
|||
| def _epoch(self, data_loader, train_model): | |||
|
|
|||
| def _epoch(self, data_loader, train_model, test_model=False): | |||
There was a problem hiding this comment.
The name test_model is misleading. It's acutally a flag to tell whether there is target values or not in the data. So I suggest using e.g. has_target=True?
Also, please add the docstring for new parameters.
There was a problem hiding this comment.
Very good suggestion, thanks !
deeprank/learn/NeuralNet.py
Outdated
| try: | ||
| self.hit_cutoff = self.state['hit_cutoff'] | ||
| except Exception: | ||
| print('No hit_cutoff provided') |
There was a problem hiding this comment.
The hit cutoff is used only if target values are provided. In prediction mode you do not need it, you need it for benchmark only.
deeprank/learn/NeuralNet.py
Outdated
| @@ -766,7 +770,8 @@ def _train(self, index_train, index_valid, index_test, | |||
| return torch.cat([param.data.view(-1) | |||
| for param in self.net.parameters()], 0) | |||
|
|
|||
| def _epoch(self, data_loader, train_model): | |||
|
|
|||
| def _epoch(self, data_loader, train_model, test_model=False): | |||
There was a problem hiding this comment.
Very good suggestion, thanks !
There was a problem hiding this comment.
Thanks @manonreau , the changes look good!
One reminder for you is that we don't use print but logger for outputing message in DeepRank :-)
Add examples of code to run the CNN models from the paper on new data:
Example with 3DeepFace
Example with the docking scoring model
Changed NeuralNet and DataSet to handle test set with no target value