def kmeans(net, layers, num_c=16, initials=None, snapshot=False, alpha=0.0):
codebook = {}
if type(num_c) == type(1):
num_c = [num_c] * len(layers)
else:
assert len(num_c) == len(layers)
print "==============Perform K-means============="
for idx, layer in enumerate(layers):
print "Eval layer:", layer
W = net.params[layer][0].data.flatten()
W = W[np.where(W != 0)]
if initials is None: # Default: uniform sample
min_W = np.min(W)
max_W = np.max(W)
initial_uni = np.linspace(min_W, max_W, num_c[idx] - 1)
codebook[layer], _ = scv.kmeans(W,
initial_uni,
compress=False,
alpha=alpha)
elif type(initials) == type(np.array([])):
codebook[layer], _ = scv.kmeans(W, initials)
elif initials == 'random':
codebook[layer], _ = scv.kmeans(W, num_c[idx] - 1)
codebook[layer] = np.append(0, codebook[layer])
print "codebook size:", len(codebook[layer])
return codebook
def kmeans_net(net, layers, num_c=16, initials=None, method='linear', compress=False):
codebook = {}
if type(num_c) == type(1):
num_c = [num_c] * len(layers)
else:
assert len(num_c) == len(layers)
print "==============Perform K-means============="
for idx, layer in enumerate(layers):
print "Eval layer:", layer
W = net.params[layer][0].data.flatten()
W = W[np.where(W != 0)]
if initials is None: # Default: uniform sample
if method == 'linear':
min_W = np.min(W)
max_W = np.max(W)
initial_uni = np.linspace(min_W, max_W, num_c[idx] - 1)
codebook[layer], _ = scv.kmeans(W, initial_uni, compress=compress)
elif method == 'random':
codebook[layer], _ = scv.kmeans(W, num_c[idx] - 1, compress=compress)
else:
raise Exception
else:
codebook[layer], _ = scv.kmeans(W, initials)
codebook[layer] = np.append(0.0, codebook[layer])
print "codebook size:", len(codebook[layer])
return codebook
def kmeans_net(net, layers, num_c=16, initials=None, snapshot=False, alpha=0.0):
codebook = {}
if type(num_c) == type(1):
num_c = [num_c] * len(layers)
else:
assert len(num_c) == len(layers)
print "==============Perform K-means============="
for idx, layer in enumerate(layers):
print "Eval layer:", layer
W = net.params[layer][0].data.flatten()
W = W[np.where(W != 0)]
if initials is None: # Default: uniform sample
min_W = np.min(W)
max_W = np.max(W)
initial_uni = np.linspace(min_W, max_W, num_c[idx] - 1)
''' # Legacy
std = np.std(W)
initial_uni = np.linspace(std * -4, std * 4, num_c[idx] - 1)
'''
if snapshot and layer=='fc6':
codebook[layer], _, codebook_history = scv.kmeans(W, initial_uni, compress=False, snapshot=True, alpha=alpha)
else:
codebook[layer], _= scv.kmeans(W, initial_uni, compress=False, alpha=alpha)
'''
codebook[layer],_= scv.kmeans(W, num_c[idx] - 1)
'''
elif type(initials) == type(np.array([])):
codebook[layer], _ = scv.kmeans(W, initials)
elif initials == 'random':
codebook[layer], _ = scv.kmeans(W, num_c[idx]-1)
codebook[layer] = np.append(0.0, codebook[layer])
print "codebook size:", len(codebook[layer])
if snapshot:
return codebook, codebook_history
return codebook
def kmeans_net(net,
layers,
num_c=16,
initials=None,
method='linear',
compress=False):
codebook = {}
if type(num_c) == type(1):
num_c = [num_c] * len(layers)
else:
assert len(num_c) == len(layers)
print "==============Perform K-means============="
for idx, layer in enumerate(layers):
print "Eval layer:", layer
W = net.params[layer][0].data.flatten()
W = W[np.where(W != 0)]
if initials is None: # Default: uniform sample
if method == 'linear':
min_W = np.min(W)
max_W = np.max(W)
initial_uni = np.linspace(min_W, max_W, num_c[idx] - 1)
codebook[layer], _ = scv.kmeans(W,
initial_uni,
compress=compress)
elif method == 'random':
codebook[layer], _ = scv.kmeans(W,
num_c[idx] - 1,
compress=compress)
else:
raise Exception
else:
codebook[layer], _ = scv.kmeans(W, initials)
codebook[layer] = np.append(0.0, codebook[layer])
print "codebook size:", len(codebook[layer])
return codebook