def learn():
labels = get_labels()
# mlr = MultiLogisticRegression(labels, conf.lr_inputs)
svm = Svm()
dataset = list(read_examples('data'))
rd.shuffle(dataset)
m = len(dataset)
train_m = int(0.7 * m)
train_set = dataset[:train_m]
test_set = dataset[train_m:]
test_m = len(test_set)
print('Learning...')
svm.learn(train_set)
print('Learnt')
correct = 0
for x, y in train_set:
pred = svm.get(x)
if pred == y:
correct += 1
print('Results on train set: {:2f}% accuracy'.format(100 * correct / train_m))
correct = 0
for x, y in test_set:
pred = svm.get(x)
if pred == y:
correct += 1
print('Results on test set: {:2f}% accuracy'.format(100 * correct / test_m))
with open('trained_data.pickle.b64', 'w') as wf:
wf.write(svm.export_model())
print('Weights written to trained_data.pickle.b64')
def recognize():
with AudioRecorder(rate=conf.sampling_rate) as recorder:
svm = Svm()
with open('trained_data.pickle.b64') as f:
svm.import_model(f.read())
while True:
print()
print('\x1b[A\x1b[2K' 'Press Enter to start recording or Ctrl+C to exit')
input()
recorder.start_recording()
print('\x1b[A\x1b[2K' '\x1b[31mREC\x1b[0m Press Enter to stop recording')
input()
data = list(recorder.bytes_to_numseq(recorder.finish_recording()))
transformed_data = list(sliding_diff(data, conf.sliding_diff_winsize))
scaled_data = scale_array(transformed_data, conf.svm_inputs)
pred = svm.get(scaled_data)
print('\x1b[A\x1b[2K' 'Predicted output: {}'.format(pred))