import numpy as np
import matplotlib.pyplot as plt
from sparse_autoencoder import SparseAutoencoder, sample_training_images
training_data = sample_training_images()[:100, :]
encoder = SparseAutoencoder(rho=0.01, L=0.0001, beta=3)
error, analytical_grad, numerical_grad = encoder.check_grad(training_data)
print "Gradient Error:", error
plt.plot(analytical_grad, numerical_grad, 'ro')
plt.plot(
[np.min(analytical_grad), np.max(analytical_grad)],
[np.min(analytical_grad), np.max(analytical_grad)], 'k-')
plt.xlabel('Analytical')
plt.ylabel('Numerical')
plt.show()
import numpy as np
import matplotlib.pyplot as plt
from sparse_autoencoder import SparseAutoencoder, sample_training_images
training_data = sample_training_images()[:100, :]
encoder = SparseAutoencoder(rho=0.01, L=0.0001, beta=3)
error, analytical_grad, numerical_grad = encoder.check_grad(training_data)
print "Gradient Error:", error
plt.plot(analytical_grad, numerical_grad, 'ro')
plt.plot([np.min(analytical_grad), np.max(analytical_grad)],
[np.min(analytical_grad), np.max(analytical_grad)], 'k-')
plt.xlabel('Analytical')
plt.ylabel('Numerical')
plt.show()
import matplotlib.pyplot as plt
from sparse_autoencoder import (SparseAutoencoder, sample_training_images,
PATCH_SIZE)
training_data = sample_training_images()
encoder = SparseAutoencoder(hidden_size=25, rho=0.01, L=0.0001, beta=3)
encoder.train(training_data)
w1 = encoder.params.w1
fig = plt.figure()
for i in xrange(w1.shape[-1]):
image = w1[:, i].reshape(*PATCH_SIZE)
ax = fig.add_subplot(5, 5, i + 1)
ax.imshow(image, cmap='binary')
plt.axis('off')
plt.show()
import matplotlib.pyplot as plt
from sparse_autoencoder import (
SparseAutoencoder, sample_training_images, PATCH_SIZE)
training_data = sample_training_images()
encoder = SparseAutoencoder(hidden_size=25, rho=0.01, L=0.0001, beta=3)
encoder.train(training_data)
w1 = encoder.params.w1
fig = plt.figure()
for i in xrange(w1.shape[-1]):
image = w1[:, i].reshape(*PATCH_SIZE)
ax = fig.add_subplot(5, 5, i+1)
ax.imshow(image, cmap='binary')
plt.axis('off')
plt.show()
def readFrames(directory, which_set, timestamps, patch_pose):
videos = []
videos = [name for name in os.listdir(directory, which_set)]
videos.sort()
num_patch = np.zeros((len(np.unique(timestamps)), 1))
i = 0
for timestamp in timestamps:
num_patch[i, 0] = np.sum((timestamps == timestamp) * 1)
i += 1
for video in videos:
frames = []
orientationFileNames = [
os.path.join(directory, video, 'Orientation', name)
for name in os.listdir(
os.path.join(directory, video, 'Orientation'))
]
tempGradientFileNames = [
os.path.join(directory, video, 'TempGradient', name)
for name in os.listdir(
os.path.join(directory, video, 'TempGradient'))
]
timeStampsFileNames = [
os.path.join(directory, video, 'TimeStamps',
name) for name in os.listdir(
os.path.join(directory, video, 'TimeStamps'))
]
orientationFileNames.sort()
tempGradientFileNames.sort()
timeStampsFileNames.sort()
for i in range(len(orientationFileNames)):
all_orientation = scio.loadmat(
orientationFileNames[i])['filteredOrientationFrame']
all_ox = scio.loadmat(tempGradientFileNames[i])['ox']
all_oy = scio.loadmat(tempGradientFileNames[i])['oy']
all_frequency = scio.loadmat(
tempGradientFileNames[i])['posLastEventPosition']
all_postimestamp = scio.loadmat(
timeStampsFileNames[i])['posTimeStamp']
frames.append(all_orientation)
frames.append(all_ox)
frames.append(all_oy)
frames.append(all_frequency)
frames.append(all_postimestamp)
n_feature = 5
inputNodes = selected_patchs(frames, num_patch, patch_pose, n_feature)
input_size = inputNodes.shape[0]
hidden_size = inputNodes / n_feature
if which_set is 'train':
model = SparseAutoencoder(input_size, hidden_size, 'Dining')
globalD = model.train(inputNodes)
if which_set is 'test':
model = SparseAutoencoder(input_size, hidden_size, 'Dining')
globalD = model.predict(inputNodes)
return inputNodes