def save_mat(net, path):
d = dict()
for l in net.layers:
if hasattr(l, 'weight'):
d[l.name + '_weight'] = data_loader.copy_to_cpu(l.weight)
d[l.name + '_bias'] = data_loader.copy_to_cpu(l.bias)
scipy.io.savemat(path, d)
def save_mat(net, path): d = dict() for l in net.layers: if hasattr(l,'weight'): d[l.name + '_weight'] = data_loader.copy_to_cpu(l.weight) d[l.name + '_bias'] = data_loader.copy_to_cpu(l.bias) scipy.io.savemat(path, d)
def get_labels(net, noisy_batches): w = data_loader.copy_to_cpu(net.layers[-2].weight) N = len(noisy_batches) * 128 y = np.zeros([w.shape[0], N], dtype = np.float32) c = 0 for b in noisy_batches: labels = data_loader.copy_to_cpu(b.labels) for i in range(len(labels)): y[int(labels[i]), i + c] = 1 c += 128 return y
def get_net_output(net, noisy_batches): w = data_loader.copy_to_cpu(net.layers[-2].weight) N = len(noisy_batches) * 128 x = np.zeros([w.shape[1], N], dtype = np.float32) c = 0 for b in noisy_batches: data, label = net.prepare_for_train(b.data, b.labels) net.fprop(data, net.output, False) net_out = data_loader.copy_to_cpu(net.outputs[-3]) x[:,c:c+128] = net_out c += 128 return x
def save_net(net, path): d = dict() d['w1'] = data_loader.copy_to_cpu(net.layers[1].weight) d['w2'] = data_loader.copy_to_cpu(net.layers[5].weight) d['w3'] = data_loader.copy_to_cpu(net.layers[9].weight) d['w4'] = data_loader.copy_to_cpu(net.layers[12].weight) d['b1'] = data_loader.copy_to_cpu(net.layers[1].bias) d['b2'] = data_loader.copy_to_cpu(net.layers[5].bias) d['b3'] = data_loader.copy_to_cpu(net.layers[9].bias) d['b4'] = data_loader.copy_to_cpu(net.layers[12].bias) if len(net.layers) == 16: d['confw'] = data_loader.copy_to_cpu(net.layers[14].weight) if hasattr(net, 'W_denoise'): d['W_denoise'] = data_loader.copy_to_cpu(net.W_denoise) scipy.io.savemat(path, d)
def save_net(net, path):
d = dict()
d['w1'] = data_loader.copy_to_cpu(net.layers[1].weight)
d['w2'] = data_loader.copy_to_cpu(net.layers[5].weight)
d['w3'] = data_loader.copy_to_cpu(net.layers[9].weight)
d['w4'] = data_loader.copy_to_cpu(net.layers[12].weight)
d['b1'] = data_loader.copy_to_cpu(net.layers[1].bias)
d['b2'] = data_loader.copy_to_cpu(net.layers[5].bias)
d['b3'] = data_loader.copy_to_cpu(net.layers[9].bias)
d['b4'] = data_loader.copy_to_cpu(net.layers[12].bias)
if len(net.layers) == 16:
d['confw'] = data_loader.copy_to_cpu(net.layers[14].weight)
if hasattr(net, 'W_denoise'):
d['W_denoise'] = data_loader.copy_to_cpu(net.W_denoise)
scipy.io.savemat(path, d)
def save(net, path):
d = dict()
d['layers'] = list()
for i in range(len(net.layers)):
l = net.layers[i]
if hasattr(l, 'weight'):
ld = dict()
ld['weight'] = data_loader.copy_to_cpu(l.weight)
ld['bias'] = data_loader.copy_to_cpu(l.bias)
ld['ind'] = i
d['layers'].append(ld)
f = open(path, 'wb')
d['stat'] = net.stat
cPickle.dump(d, f, protocol=-1)
f.close()
def save(net, path): d = dict() d['layers'] = list() for i in range(len(net.layers)): l = net.layers[i] if hasattr(l,'weight'): ld = dict() ld['weight'] = data_loader.copy_to_cpu(l.weight) ld['bias'] = data_loader.copy_to_cpu(l.bias) ld['ind'] = i d['layers'].append(ld) f = open(path, 'wb') d['stat'] = net.stat cPickle.dump(d, f, protocol=-1) f.close()
def save_net(net, path): d = dict() d['w1'] = data_loader.copy_to_cpu(net.layers[1].weight) d['w2'] = data_loader.copy_to_cpu(net.layers[5].weight) d['w3'] = data_loader.copy_to_cpu(net.layers[9].weight) d['w4'] = data_loader.copy_to_cpu(net.layers[12].weight) d['b1'] = data_loader.copy_to_cpu(net.layers[1].bias) d['b2'] = data_loader.copy_to_cpu(net.layers[5].bias) d['b3'] = data_loader.copy_to_cpu(net.layers[9].bias) d['b4'] = data_loader.copy_to_cpu(net.layers[12].bias) scipy.io.savemat(path, d)
def get_scores(net, batches): batch_size = batches[0].data.shape[1] N = len(batches) * batch_size score = np.zeros(N) i = 0 for batch in batches: net.train_batch(batch.data, batch.labels, TEST) score[i:i+batch_size] = -1 * data_loader.copy_to_cpu(net.layers[-1].cost).reshape(batch_size) i += batch_size return score
def get_scores(net, batches):
batch_size = batches[0].data.shape[1]
N = len(batches) * batch_size
score = np.zeros(N)
i = 0
for batch in batches:
net.train_batch(batch.data, batch.labels, TEST)
score[i:i + batch_size] = -1 * data_loader.copy_to_cpu(
net.layers[-1].cost).reshape(batch_size)
i += batch_size
return score
def show_stat(net, test_batches):
plt.clf()
f = plt.gcf()
c = confusion_matrix.get_confusion(net, test_batches)
f.add_subplot('221')
plt.imshow(c, interpolation = 'nearest')
plt.colorbar()
f.add_subplot('222')
m = data_loader.copy_to_cpu(net.layers[-2].weight)
plt.imshow(m, interpolation = 'nearest')
plt.colorbar()
f.add_subplot('223')
plt.plot(net.stat['test-error'])
plt.draw()
time.sleep(0.05)
def show_stat(net, test_batches):
plt.clf()
f = plt.gcf()
c = confusion_matrix.get_confusion(net, test_batches)
f.add_subplot('221')
plt.imshow(c, interpolation='nearest')
plt.colorbar()
f.add_subplot('222')
m = data_loader.copy_to_cpu(net.layers[-2].weight)
plt.imshow(m, interpolation='nearest')
plt.colorbar()
f.add_subplot('223')
plt.plot(net.stat['test-error'])
plt.draw()
time.sleep(0.05)
def update_confusion_matrix(net, noisy_batches, y, wc): w = data_loader.copy_to_cpu(net.layers[-2].weight) x = get_net_output(net, noisy_batches) w = train(w, x, y, 1, wc, 50) w = train(w, x, y, 0.1, wc, 10) net.layers[-2].weight = data_loader.copy_to_gpu(w)