def testGenerateMultidimAccuracy(self):
'''Test the accuracy of generation of features with multidim generators'''
data = np.array([[[1,2,3],[1,2,3],[1,2,3]]])
gs = [[gens.minimum], [gens.maximum], [gens.mean]]
fg = FeatureGenerator(data, gs)
fg.generate()
expected_genfeat = np.array([[1,3,2]])
assert np.shape(fg.generated_features) == np.shape(expected_genfeat), 'Generate features produced an incorrect number of features'
assert (fg.generated_features == expected_genfeat).all(), 'Generated features are not correct'
def testGenerateMultidimAccuracyDiff2ndDim(self):
'''Test the accuracy of generation of features with multidim generators with oddly shaped generator dimensions'''
data = np.array([[[5,4,5],[1,2,3],[0,-1,-2]]])
gs = [[gens.minimum], [gens.maximum, gens.mean, gens.minimum], [gens.mean]]
fg = FeatureGenerator(data, gs)
fg.generate()
expected_genfeat = np.array([[4,3,2,1,-1]])
assert np.shape(fg.generated_features) == np.shape(expected_genfeat), 'Generate features produced an incorrect number of features'
assert (fg.generated_features == expected_genfeat).all(), 'Generated features are not correct'
def testSelect(self):
'''Test the selection of generated features'''
data = np.random.rand(200,3,10)
fg = FeatureGenerator(data)
fg.generators = [gens.identity]
fg.generate()
selret = fg.select()
assert len(fg.selected_weights) > 0, 'Bad selected weights output'
assert len(fg.selected_fracs) > 0, 'Bad selected fracs output'
assert (selret == fg.selected_fracs).all(), 'Select() did not return selected fracs appropriately'
assert len(fg.selected_fracs) == fg.selected_weights.shape[0], \
'Selected fracs and weights do not have equal first dimension'
def testMultiDimGeneratorsWithNones(self):
'''Test the generation of features with multidim generators with none generators'''
data = np.random.rand(200,3,10)
gs = [[], [gens.maximum], []]
fg = FeatureGenerator(data, gs)
fg.generate()
assert len(fg.generated_features) > 0, 'Bad generate() output'
assert len(fg.generated_features.shape) == 2, 'Generated features is not 2-d'
assert fg.generated_features.shape[0] == fg.datas.shape[0], \
'Number of observations is not held constant during generate()'
assert fg.generated_features.shape[1] == 1, \
'Number of generated features is not appropriate'
def testGenerateDiff2ndDim(self):
'''Test the generation of features with multidim generators with oddly shaped generator dimensions'''
data = np.random.rand(200,3,10)
gs = [[gens.minimum], [gens.maximum, gens.mean], [gens.mean, gens.identity]]
fg = FeatureGenerator(data, gs)
fg.generated_features = [] # Reset the generator
fg.generate()
assert len(fg.generated_features) > 0, 'Bad generate() output'
assert len(fg.generated_features.shape) == 2, 'Generated features is not 2-d'
assert fg.generated_features.shape[0] == fg.datas.shape[0], \
'Number of observations is not held constant during generate()'
assert fg.generated_features.shape[1] == 14, \
'Number of generated features is not appropriate'
def testSelectThreshold(self):
'''Test the effect of the threshold parameter on feature selection'''
data = np.random.rand(200,3,10)
fg = FeatureGenerator(data)
fg.generators = [gens.identity]
fg.generate()
fg.select(0.00001)
n01 = len(fg.selected_fracs)
fg.select(0.5)
n5 = len(fg.selected_fracs)
sum5 = sum(fg.selected_fracs)
assert n5 <= n01, 'Threshold has no effect on select()'
assert sum5 < .75, 'Threshold does not bound eigenvalues correctly'
def testGenerateScaling(self):
'''Test feature scaling'''
data = np.array([[np.arange(10),np.arange(10)-5]],dtype='float')
params = dict({'scalaing':False})
generators = [gens.identity]
fg = FeatureGenerator(data, generators=generators, gen_params=params)
fg.generate()
assert len(fg.scaling[0]) > 0, 'Scaling not generated'
assert len(fg.scaling[0]) == len(fg.scaling[1]), 'Scaling parameters sized inappropriately'
assert len(fg.scaling[0]) == fg.datas.shape[1], 'Scaling parameters sized incorrectly'
assert fg.scaling[0].max() == 1 and fg.scaling[0].min() == 1, \
'Scaling "off" did not produce correct multipliers'
assert fg.scaling[1].max() == 0 and fg.scaling[1].min() == 0, \
'Scaling "off" did not produce correct summands'
assert fg.generated_features.max() > 1, 'Scaling applied when off'
fg.generator_params['scaling'] = True
fg.generate()
assert len(fg.scaling[0]) > 0, 'Scaling not generated with scaling on'
assert len(fg.scaling[0]) == len(fg.scaling[1]), 'Scaling parameters sized inappropriately with scaling on'
assert len(fg.scaling[0]) == fg.datas.shape[1], 'Scaling parameters sized incorrectly with scaling on'
assert not (fg.scaling[0].max() == 1 and fg.scaling[0].min() == 1), \
'Scaling "on" did not produce correct multipliers'
assert not (fg.scaling[1].max() == 0 and fg.scaling[1].min() == 0), \
'Scaling "on" did not produce correct summands'
assert fg.generated_features.max() == 1, 'Scaling (maximum) not applied correctly'
assert fg.generated_features.min() == -1, 'Scaling (minimum) not applied correctly'
assert fg.generated_features[0,0:10].min() == 0, 'Scaling on a per vector basis not applied correctly'
def testGenerate(self):
'''Test the generation of features using the basic generators'''
data = np.random.rand(200,3,10)
fg = FeatureGenerator(data)
gs = [gens.identity, gens.minimum, gens.maximum, gens.mean, gens.stdev]
fg.generators = gs;
fg.generated_features = [] # Reset the generator
fg.generate()
assert len(fg.generated_features) > 0, 'Bad generate() output'
assert len(fg.generated_features.shape) == 2, 'Generated features is not 2-d'
assert fg.generated_features.shape[0] == fg.datas.shape[0], \
'Number of observations is not held constant during generate()'
fg.generators = [gens.identity, gens.maximum, gens.minimum, gens.mean]
fg.datas = np.array([[[1,2,3]]])
fg.generate()
expected_genfeat = np.array([[1,2,3,3,1,2]])
assert (fg.generated_features == expected_genfeat).all(), 'Generated features are not aligned correctly'
class GeneratorTestCase(unittest.TestCase):
def setUp(self):
data = np.random.rand(200,3,10)
self.fg = FeatureGenerator(data)
def testDwtGenerator(self):
'''Test the use of the dwt generator'''
# Run as default
self.fg.generators = [gens.identity, gens.minimum, gens.dwt]
self.fg.generate()
# Test the use of a bad wavelet
self.fg.generator_params['dwt:wavelet'] = 'wrong'
expected_error = False
try:
self.fg.generate()
except ValueError:
expected_error = True
assert expected_error, 'Bad wavelet parameter did not produce error'
# Test the use of a non-default wavelet
self.fg.generator_params['dwt:wavelet'] = 'haar'
self.fg.generate()
assert len(self.fg.generated_features) > 0, 'Bad generate() output with dwt'
assert len(self.fg.generated_features.shape) == 2, 'Generated features is not 2-d'
assert self.fg.generated_features.shape[0] == self.fg.datas.shape[0], \
'Number of observations is not held constant during generate()'
def testFftGenerator(self):
'''Test the use of the fft generator'''
# Run as default
self.fg.generators = [gens.identity, gens.minimum, gens.fft]
self.fg.generate()
assert len(self.fg.generated_features) > 0, 'Bad generate() output with fft'
assert len(self.fg.generated_features.shape) == 2, 'Generated features is not 2-d'
assert self.fg.generated_features.shape[0] == self.fg.datas.shape[0], \
'Number of observations is not held constant during generate()'
def testApplication(self):
'''Test the application of the generated weights on data'''
data = np.random.rand(200,3,10)
fg = FeatureGenerator(data)
fg.generators = [gens.identity]
fg.generate()
fg.select()
applied_base = fg.apply_weights()
assert len(applied_base.shape) == 2, 'apply_weights() returns non-2d shape for [] input'
assert applied_base.shape[0] == fg.datas.shape[0], \
'apply_weights() doesnt keep first dimension constant on [] input'
assert applied_base.shape[1] == len(fg.selected_fracs), \
'apply_weights() does not keep the same number of features as selected'
new_data = np.random.rand(int(fg.datas.shape[0]/2), fg.datas.shape[1], fg.datas.shape[2])
applied_rand = fg.apply_weights(new_data)
assert len(applied_rand.shape) == 2, 'apply_weights() returns non-2d shape for new data input'
assert applied_rand.shape[0] == new_data.shape[0], \
'apply_weights() doesnt keep first dimension constant on new data input'
assert applied_rand.shape != applied_base.shape, 'New input data had no effect on return from apply_weights()'
def setUp(self):
data = np.random.rand(200,3,10)
self.fg = FeatureGenerator(data)
