def main():
"""
"""
import random
import numpy
recogniter = ClaClassifier(net_structure, sensor_params, dest_resgion_data, class_encoder_params)
print "training ..."
for i, data in enumerate(train_data[:1000]):
patch_data, movement = get_patch(data, height=patch_heigh, width=patch_width, step=patch_step)
for patch in patch_data:
input_len = reduce(lambda x, y: x * y, patch.shape)
input_data = {"pixel": patch.reshape((input_len)).tolist(), "label": train_label[i][0]}
inferences = recogniter.run(input_data, learn=True, class_learn=True, learn_layer=None)
# recogniter.layer_output(input_data)
# recogniter.print_inferences(input_data, inferences)
# print train_label[i][0] , inferences['classifier_region1']['best']
recogniter.reset()
# validate
if i % 50 == 0 and not i == 0:
valid = validate(recogniter, test_data, test_label, limit=30)
print "%d : valid: %8.5f" % (i, valid)
def validate(model, test_data, test_label, limit=100):
result = []
tdata = test_data[:limit]
for i, data in enumerate(tdata):
if test_label[i][0] not in (0, 1):
continue
patch_result = Counter()
patch_data, movement = get_patch(data, height=patch_heigh, width=patch_width, step=patch_step)
for patch in patch_data:
input_len = reduce(lambda x,y: x * y, patch.shape)
input_data = {
'pixel': patch.reshape((input_len)).tolist() ,
'label': 'no'
}
model.disableLearning()
modelresult = model.run(input_data)
#print label[i][0], result.inferences['multiStepBestPredictions']
best_result = modelresult.inferences['multiStepBestPredictions']
#patch_result[best_result['value']] += best_result['prob']
patch_result[best_result[0]] += 1
# print test_label[i][0]
# print patch_result
if test_label[i][0] == max(patch_result.items(), key=lambda x:x[1])[0]:
result.append(1)
#model.resetSequenceStates()
model._getTPRegion().getSelf().resetSequenceStates()
return len(result)/len(tdata)
def validate(recogniter, test_data, test_label, limit=100):
result = []
tdata = test_data[:limit]
for i, data in enumerate(tdata):
patch_result = Counter()
patch_data, movement = get_patch(data, height=patch_heigh, width=patch_width, step=patch_step)
for patch in patch_data:
input_len = reduce(lambda x, y: x * y, patch.shape)
input_data = {"pixel": patch.reshape((input_len)).tolist(), "label": "no"}
inferences = recogniter.run(input_data, learn=True, class_learn=False, learn_layer=None)
best_result = inferences["classifier_region1"]["best"]
# patch_result[best_result['value']] += best_result['prob']
patch_result[best_result["value"]] += 1
# print test_label[i][0]
# print patch_result
if test_label[i][0] == max(patch_result.items(), key=lambda x: x[1])[0]:
result.append(1)
recogniter.reset()
return float(len(result)) / len(tdata)
def validate(recogniter, test_data, test_label, limit=100):
result = []
tdata = test_data[:limit]
for i, data in enumerate(tdata):
patch_result = Counter()
patch_data, movement = get_patch(data,
height=patch_heigh,
width=patch_width,
step=patch_step)
for patch in patch_data:
input_len = reduce(lambda x, y: x * y, patch.shape)
input_data = {
'pixel': patch.reshape((input_len)).tolist(),
'label': 'no'
}
inferences = recogniter.run(input_data,
learn=True,
class_learn=False,
learn_layer=None)
best_result = inferences['classifier_region1']['best']
#patch_result[best_result['value']] += best_result['prob']
patch_result[best_result['value']] += 1
# print test_label[i][0]
# print patch_result
if test_label[i][0] == max(patch_result.items(),
key=lambda x: x[1])[0]:
result.append(1)
recogniter.reset()
return float(len(result)) / len(tdata)
def test_get_patch_step_2(self):
# step = 2
patch_data, movement = get_patch(self.tdata, 2, 2, 2)
self.assertEqual(patch_data.shape, (1, 2, 2, 3))
self.assertEqual(patch_data.tolist()[0][0][0], [7., 8., 9.])
self.assertEqual(patch_data.tolist()[0][0][1], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[0][1][0], [4., 5., 6.])
self.assertEqual(patch_data.tolist()[0][1][1], [5., 6., 7.])
def test_get_patch_step_2(self):
# step = 2
patch_data, movement = get_patch(self.tdata, 2, 2, 2)
self.assertEqual(patch_data.shape, (1,2,2,3))
self.assertEqual(patch_data.tolist()[0][0][0], [7., 8., 9.] )
self.assertEqual(patch_data.tolist()[0][0][1], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[0][1][0], [4., 5., 6.] )
self.assertEqual(patch_data.tolist()[0][1][1], [5., 6., 7.] )
def test_get_patch(self):
# [1, 1]
patch_data, movement = get_patch(self.tdata, 1, 1, 1)
self.assertEqual(patch_data.shape, (9, 1, 1, 3))
self.assertEqual(patch_data.tolist()[0][0][0], [7., 8., 9.])
self.assertEqual(patch_data.tolist()[1][0][0], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[2][0][0], [0., 0., 0.])
self.assertEqual(patch_data.tolist()[3][0][0], [0., 0., 0.])
self.assertEqual(patch_data.tolist()[4][0][0], [5., 6., 7.])
def test_get_patch(self):
# [1, 1]
patch_data, movement = get_patch(self.tdata, 1, 1, 1)
self.assertEqual(patch_data.shape, (9,1,1,3))
self.assertEqual(patch_data.tolist()[0][0][0], [7., 8., 9.] )
self.assertEqual(patch_data.tolist()[1][0][0], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[2][0][0], [0., 0., 0.] )
self.assertEqual(patch_data.tolist()[3][0][0], [0., 0., 0.] )
self.assertEqual(patch_data.tolist()[4][0][0], [5., 6., 7.] )
def test_get_patch_form_2(self):
# [2, 2]
patch_data, movement = get_patch(self.tdata, 2, 2, 1)
self.assertEqual(patch_data.shape, (4, 2, 2, 3))
self.assertEqual(patch_data.tolist()[0][0][0], [7., 8., 9.])
self.assertEqual(patch_data.tolist()[0][0][1], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[0][1][0], [4., 5., 6.])
self.assertEqual(patch_data.tolist()[0][1][1], [5., 6., 7.])
self.assertEqual(patch_data.tolist()[1][0][0], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[1][0][1], [0., 0., 0.])
self.assertEqual(patch_data.tolist()[1][1][0], [5., 6., 7.])
self.assertEqual(patch_data.tolist()[1][1][1], [0., 0., 0.])
def test_get_patch_form_2(self):
# [2, 2]
patch_data, movement = get_patch(self.tdata, 2, 2, 1)
self.assertEqual(patch_data.shape, (4,2,2,3))
self.assertEqual(patch_data.tolist()[0][0][0], [7., 8., 9.] )
self.assertEqual(patch_data.tolist()[0][0][1], [8., 9., 10.])
self.assertEqual(patch_data.tolist()[0][1][0], [4., 5., 6.] )
self.assertEqual(patch_data.tolist()[0][1][1], [5., 6., 7.] )
self.assertEqual(patch_data.tolist()[1][0][0], [8., 9., 10.] )
self.assertEqual(patch_data.tolist()[1][0][1], [0., 0., 0.])
self.assertEqual(patch_data.tolist()[1][1][0], [5., 6., 7.] )
self.assertEqual(patch_data.tolist()[1][1][1], [0., 0., 0.] )
def main():
"""
"""
import random
import numpy
model = createModel()
shifter = InferenceShifter()
tobological_data, label = load_dataset(
'./data/pylearn2_gcn_whitened/train.pkl')
for i, data in enumerate(tobological_data[:1000]):
if label[i][0] not in (0, 1):
continue
patch_data, movement = get_patch(data,
height=patch_heigh,
width=patch_width,
step=patch_step)
print '%d, label:%s, ' % (i, label[i][0]),
for data in patch_data:
input_len = reduce(lambda x, y: x * y, data.shape)
input_data = {
'pixel': data.reshape((input_len)).tolist(),
'label': label[i][0]
}
model.enableLearning()
result = model.run(input_data)
#result = shifter.shift(result)
print label[i][0], result.inferences['multiStepBestPredictions']
#model.resetSequenceStates()
#model._getTPRegion().executeCommand(['resetSequenceStates'])
#model._getTPRegion().resetSequenceStates()
model._getTPRegion().getSelf().resetSequenceStates()
# validate
if i % 3 == 0:
valid = validate(model, test_data, test_label, limit=30)
print '%d : valid: %8.5f' % (i, valid)
def main():
"""
"""
import random
import numpy
model = createModel()
shifter = InferenceShifter()
tobological_data, label = load_dataset('./data/pylearn2_gcn_whitened/train.pkl')
for i, data in enumerate(tobological_data[:1000]):
if label[i][0] not in (0, 1):
continue
patch_data, movement = get_patch(data, height=patch_heigh, width=patch_width, step=patch_step)
print '%d, label:%s, ' % (i, label[i][0]),
for data in patch_data:
input_len = reduce(lambda x,y: x * y, data.shape)
input_data = {
'pixel': data.reshape((input_len)).tolist() ,
'label': label[i][0]
}
model.enableLearning()
result = model.run(input_data)
#result = shifter.shift(result)
print label[i][0], result.inferences['multiStepBestPredictions']
#model.resetSequenceStates()
#model._getTPRegion().executeCommand(['resetSequenceStates'])
#model._getTPRegion().resetSequenceStates()
model._getTPRegion().getSelf().resetSequenceStates()
# validate
if i % 3 == 0:
valid = validate(model, test_data, test_label, limit=30)
print '%d : valid: %8.5f' % (i, valid)
def main():
"""
"""
import random
import numpy
recogniter = ClaClassifier(net_structure, sensor_params, dest_resgion_data,
class_encoder_params)
print 'training ...'
for i, data in enumerate(train_data[:1000]):
patch_data, movement = get_patch(data,
height=patch_heigh,
width=patch_width,
step=patch_step)
for patch in patch_data:
input_len = reduce(lambda x, y: x * y, patch.shape)
input_data = {
'pixel': patch.reshape((input_len)).tolist(),
'label': train_label[i][0]
}
inferences = recogniter.run(input_data,
learn=True,
class_learn=True,
learn_layer=None)
#recogniter.layer_output(input_data)
#recogniter.print_inferences(input_data, inferences)
#print train_label[i][0] , inferences['classifier_region1']['best']
recogniter.reset()
# validate
if i % 50 == 0 and not i == 0:
valid = validate(recogniter, test_data, test_label, limit=30)
print '%d : valid: %8.5f' % (i, valid)
def validate(model, test_data, test_label, limit=100):
result = []
tdata = test_data[:limit]
for i, data in enumerate(tdata):
if test_label[i][0] not in (0, 1):
continue
patch_result = Counter()
patch_data, movement = get_patch(data,
height=patch_heigh,
width=patch_width,
step=patch_step)
for patch in patch_data:
input_len = reduce(lambda x, y: x * y, patch.shape)
input_data = {
'pixel': patch.reshape((input_len)).tolist(),
'label': 'no'
}
model.disableLearning()
modelresult = model.run(input_data)
#print label[i][0], result.inferences['multiStepBestPredictions']
best_result = modelresult.inferences['multiStepBestPredictions']
#patch_result[best_result['value']] += best_result['prob']
patch_result[best_result[0]] += 1
# print test_label[i][0]
# print patch_result
if test_label[i][0] == max(patch_result.items(),
key=lambda x: x[1])[0]:
result.append(1)
#model.resetSequenceStates()
model._getTPRegion().getSelf().resetSequenceStates()
return len(result) / len(tdata)