def g_error(pic, w1, w2, ans, N):
data = np.zeros((6, 36))
a = np.zeros((6, 2))
for i in range(0, 6):
if random.randint(0, 1) == 1:
p = p_gn(N)
p1 = tf.reshape(p.astype('float64').reshape(1, 3000) / 255,
shape=[-1, 50, 60, 1])
II = conv2d(p1, w1)
II = maxpool2d(II, k=5)
II1 = conv2d(II, w2)
II1 = maxpool2d(II1, k=5)
II1 = tf.reshape(II1, [-1, 36]).eval()
data[i, :] = II1
a[i, :] = [1, 0]
if random.randint(0, 1) == 0:
k = random.randint(0, 5)
while k == ans:
k = random.randint(0, 5)
p = pic[:, :, k]
p1 = tf.reshape(p.astype('float64').reshape(1, 3000) / 255,
shape=[-1, 50, 60, 1])
II = conv2d(p1, w1)
II = maxpool2d(II, k=5)
II1 = conv2d(II, w2)
II1 = maxpool2d(II1, k=5)
II1 = tf.reshape(II1, [-1, 36]).eval()
data[i, :] = II1
a[i, :] = [0, 1]
return data, a
def forward(p, pic, w1, w2, w, b):
with tf.Session() as sess:
p1 = tf.reshape(p.astype('float64').reshape(1, 3000) / 255,
shape=[-1, 50, 60, 1])
I = conv2d(p1, w1)
I = maxpool2d(I, k=5)
I1 = conv2d(I, w2)
I1 = maxpool2d(I1, k=5)
I1 = tf.reshape(I1, [-1, 36]).eval()
II2 = np.zeros((6, 36))
with tf.Session() as sess:
for i in range(0, 6):
p1 = tf.reshape(pic[:, :, i].astype('float64').reshape(1, 3000) /
255,
shape=[-1, 50, 60, 1])
II = conv2d(p1, w1)
II = maxpool2d(II, k=5)
II1 = conv2d(II, w2)
II1 = maxpool2d(II1, k=5)
II1 = tf.reshape(II1, [-1, 36]).eval()
II2[i, :] = II1
II3 = np.zeros((6, 1024))
II1 = I1.dot(w) + b
with tf.Session() as sess:
for i in range(0, 6):
II3[i, :] = II2[i, :].dot(w) + b
return II1, II3
def forward(pic,w1,w2):
II=np.zeros((16,6*7*7))
with tf.Session() as sess:
for i in range (0,16):
p1=tf.reshape(pic[:,:,i].astype('float64').reshape(1,28*28)/255 , shape=[-1, 28,28 , 1])
I=conv2d(p1, w1)
I=maxpool2d(I,k=2)
I1 = conv2d(I, w2)
I1=maxpool2d(I1,k=2)
I1=tf.reshape(I1,[-1,6*7*7]).eval()
II[i,:]=I1
return II
def forward(pic, w1, w2):
II = np.zeros((9, 180))
with tf.Session() as sess:
for i in range(0, 9):
p1 = tf.reshape(
pic[:, :, i].astype('float64').reshape(1, 40 * 50) / 255,
shape=[-1, 40, 50, 1])
I = conv2d(p1, w1)
I = maxpool2d(I, k=3)
I1 = conv2d(I, w2)
I1 = maxpool2d(I1, k=3)
I1 = tf.reshape(I1, [-1, 180]).eval()
II[i, :] = I1
return II
def prep(pic, weight1, weight2):
n = weight1.shape[4]
M = np.zeros((n, pic.shape[2], 7020))
with tf.Session() as sess:
p = np.zeros((pic.shape[2], 3000))
for i in range(0, pic.shape[2]):
p[i, :] = pic[:, :, i].reshape(3000, )
p = tf.reshape(p, shape=[-1, 50, 60, 1])
for k in range(0, n):
I = conv2d(p, weight1[:, :, :, :, k].reshape(25, 25, 1, 3))
I = maxpool2d(I, k=2)
I1 = conv2d(I, weight2[:, :, :, :, k].reshape(50, 50, 3, 6))
# Max Pooling (down-sampling)
I1 = maxpool2d(I1, k=2)
I1 = tf.reshape(I1, [-1, 7020]).eval()
M[k, :, :] = I1
return M
def prep(pic, weight1, weight2, a):
n = weight1.shape[4]
M = np.zeros((n, pic.shape[2], 49 * 2 * a))
with tf.Session() as sess:
p = np.zeros((pic.shape[2], 784))
for i in range(0, pic.shape[2]):
p[i, :] = pic[:, :, i].reshape(784, )
p = tf.reshape(p, shape=[-1, 28, 28, 1])
for k in range(0, n):
I = conv2d(p, weight1[:, :, :, :, k].reshape(5, 5, 1, a))
I = maxpool2d(I, 2)
I1 = conv2d(I, weight2[:, :, :, :, k].reshape(5, 5, a, 2 * a))
# Max Pooling (down-sampling)
I1 = maxpool2d(I1, 2)
I1 = tf.reshape(I1, [-1, 49 * 2 * a]).eval()
M[k, :, :] = I1
return M
def forward(p, pic, w1, w2):
with tf.Session() as sess:
p1 = tf.reshape(p.astype('float64').reshape(1, 3000) / 255,
shape=[-1, 50, 60, 1])
I = conv2d(p1, w1)
I = maxpool2d(I, k=5)
#I1 = conv2d(I, w2)
#I1=maxpool2d(I1,k=5)
I1 = tf.reshape(I, [-1, 360]).eval()
II2 = np.zeros((6, 360))
with tf.Session() as sess:
for i in range(0, 6):
p1 = tf.reshape(pic[:, :, i].astype('float64').reshape(1, 3000) /
255,
shape=[-1, 50, 60, 1])
II = conv2d(p1, w1)
II = maxpool2d(II, k=5)
#II1 = conv2d(II, w2)
#II1=maxpool2d(II1,k=5)
II1 = tf.reshape(II, [-1, 360]).eval()
II2[i, :] = II1
return I1, II2
