WIP: Add annotation support and fix some bugs
Merge request reports
Activity
added 1 commit
- 73173c32 - Allow for missing files in the projection step
added 1 commit
- 30b6e0a6 - Add support for training using the development data too
added 1 commit
- ce0a6a66 - Dev files for training should not be combined
assigned to @pkorshunov
32 32 parser = parsers['main'] 33 33 parser.add_argument('--sub-task', 34 34 choices=('preprocess', 'train-extractor', 'extract', 'train-projector', 'project', 35 'compute-scores'), 35 'train', 'compute-scores'), @amohammadi What's the point of adding another 'train' task? This looks redundant. Algorithm is trained using 'train-projector', similar to
bob.bio.base
.Think of it like this: you want to do PCA+SVM, this means that you project the features (PCA) and then train an algorithm. This is possible in bob.bio with
train_projector
,train_enroller
, andenroll
but no such thing is available here.Edited by Amir MOHAMMADI
40 40 # now that we have set up everything, get the command line arguments 41 41 return tools.initialize(parsers, command_line_parameters, 42 42 skips=['preprocessing', 'extractor-training', 'extraction', 'projector-training', 43 'projection', 'score-computation']) 43 'projection', 'algorithm-training', 'score-computation']) @amohammadi The same as above: 'algorithm-training'=='projector-training'
211 237 The file to read the projector from. 212 238 """ 213 239 pass 240 241 def train(real, attack, model_file): @amohammadi So, this method is redundant as well.
18 18 assert (data in _data) 19 19 return data + 1.0 20 20 21 def train(self, training_data, extractor_file): @amohammadi Not sure why this function is here at all. It is optional for Extractor.
Please register or sign in to reply