diff --git a/bob/bio/base/extractor/stacks.py b/bob/bio/base/extractor/stacks.py index 8f4af48f8d304c35bee817169652b961885df430..6bc3cb5f452e03bb879ebcf2453c7aefb78c92c5 100644 --- a/bob/bio/base/extractor/stacks.py +++ b/bob/bio/base/extractor/stacks.py @@ -33,7 +33,7 @@ class MultipleExtractor(Extractor): training_data = [e(d) for d in training_data] # if any of the extractors require splitting the data, the # split_training_data_by_client is True. - if e.split_training_data_by_client: + elif e.split_training_data_by_client: e.train(training_data, extractor_file) if not apply: return @@ -62,7 +62,8 @@ class MultipleExtractor(Extractor): groups = self.get_extractor_groups() for e, group in zip(self.processors, groups): f.cd(group) - e.load(f) + if e.requires_training: + e.load(f) f.cd('..') @@ -110,10 +111,15 @@ class SequentialExtractor(SequentialProcessor, MultipleExtractor): def train(self, training_data, extractor_file): with HDF5File(extractor_file, 'w') as f: groups = self.get_extractor_groups() - for e, group in zip(self.processors, groups): + for i, (e, group) in enumerate(zip(self.processors, groups)): + if i == len(self.processors) - 1: + apply = False + else: + apply = True f.create_group(group) f.cd(group) - training_data = self.train_one(e, training_data, f, apply=True) + training_data = self.train_one(e, training_data, f, + apply=apply) f.cd('..') def read_feature(self, feature_file): diff --git a/bob/bio/base/test/dummy/extractor.py b/bob/bio/base/test/dummy/extractor.py index a3aaf6f7ea04347393db8cbc8efd1dac95e98000..eca7517ca8a46676074f93410d95b98d71757721 100644 --- a/bob/bio/base/test/dummy/extractor.py +++ b/bob/bio/base/test/dummy/extractor.py @@ -1,5 +1,5 @@ import numpy -import bob.io.base +import bob.bio.base from bob.bio.base.extractor import Extractor @@ -12,10 +12,10 @@ class DummyExtractor (Extractor): def train(self, train_data, extractor_file): assert isinstance(train_data, list) - bob.io.base.save(_data, extractor_file) + bob.bio.base.save(_data, extractor_file) def load(self, extractor_file): - data = bob.io.base.load(extractor_file) + data = bob.bio.base.load(extractor_file) assert (_data == data).all() self.model = True diff --git a/bob/bio/base/test/test_stacks.py b/bob/bio/base/test/test_stacks.py index 926901382af5455ee292ace46abc885582baaaa6..b1abac3c0648a6fcd95b46e0150a7318c0242909 100644 --- a/bob/bio/base/test/test_stacks.py +++ b/bob/bio/base/test/test_stacks.py @@ -1,11 +1,13 @@ from functools import partial import numpy as np +import tempfile from bob.bio.base.utils.processors import ( SequentialProcessor, ParallelProcessor) from bob.bio.base.preprocessor import ( SequentialPreprocessor, ParallelPreprocessor, CallablePreprocessor) from bob.bio.base.extractor import ( SequentialExtractor, ParallelExtractor, CallableExtractor) +from bob.bio.base.test.dummy.extractor import extractor as dummy_extractor DATA = [0, 1, 2, 3, 4] PROCESSORS = [partial(np.power, 2), np.mean] @@ -43,3 +45,13 @@ def test_extractors(): proc = ParallelExtractor(processors) data = proc(DATA) assert all(np.allclose(x1, x2) for x1, x2 in zip(data, PAR_DATA)) + + +def test_trainable_extractors(): + processors = [CallableExtractor(p) for p in PROCESSORS] + [dummy_extractor] + proc = SequentialExtractor(processors) + with tempfile.NamedTemporaryFile(suffix='.hdf5') as f: + proc.train(DATA, f.name) + proc.load(f.name) + data = proc(DATA) + assert np.allclose(data, SEQ_DATA)