In [1]:
import numpy as np
from keras.models import Model
from keras.layers import Input
from keras.layers.convolutional import SeparableConv2D
from keras import backend as K
import json
from collections import OrderedDict
In [2]:
def format_decimal(arr, places=6):
return [round(x * 10**places) / 10**places for x in arr]
In [3]:
DATA = OrderedDict()
[convolutional.SeparableConv2D.0] 4 3x3 filters on 5x5x2 input, strides=(1,1), padding='valid', data_format='channels_last', depth_multiplier=1, activation='linear', use_bias=True
In [4]:
data_in_shape = (5, 5, 2)
conv = SeparableConv2D(4, (3,3), strides=(1,1),
padding='valid', data_format='channels_last',
depth_multiplier=1, activation='linear', use_bias=True)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(160)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
print('b shape:', weights[2].shape)
print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.0'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
[convolutional.SeparableConv2D.1] 4 3x3 filters on 5x5x2 input, strides=(1,1), padding='valid', data_format='channels_last', depth_multiplier=2, activation='relu', use_bias=True
In [5]:
data_in_shape = (5, 5, 2)
conv = SeparableConv2D(4, (3,3), strides=(1,1),
padding='valid', data_format='channels_last',
depth_multiplier=2, activation='relu', use_bias=True)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(161)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
print('b shape:', weights[2].shape)
print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.1'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
[convolutional.SeparableConv2D.2] 16 3x3 filters on 5x5x4 input, strides=(1,1), padding='valid', data_format='channels_last', depth_multiplier=3, activation='relu', use_bias=True
In [6]:
data_in_shape = (5, 5, 4)
conv = SeparableConv2D(16, (3,3), strides=(1,1),
padding='valid', data_format='channels_last',
depth_multiplier=3, activation='relu', use_bias=True)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(162)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
print('b shape:', weights[2].shape)
print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.2'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
[convolutional.SeparableConv2D.3] 4 3x3 filters on 5x5x2 input, strides=(2,2), padding='valid', data_format='channels_last', depth_multiplier=1, activation='relu', use_bias=True
In [7]:
data_in_shape = (5, 5, 2)
conv = SeparableConv2D(4, (3,3), strides=(2,2),
padding='valid', data_format='channels_last',
depth_multiplier=1, activation='relu', use_bias=True)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(163)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
print('b shape:', weights[2].shape)
print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.3'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
[convolutional.SeparableConv2D.4] 4 3x3 filters on 5x5x2 input, strides=(1,1), padding='same', data_format='channels_last', depth_multiplier=1, activation='relu', use_bias=True
In [8]:
data_in_shape = (5, 5, 2)
conv = SeparableConv2D(4, (3,3), strides=(1,1),
padding='same', data_format='channels_last',
depth_multiplier=1, activation='relu', use_bias=True)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(164)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
print('b shape:', weights[2].shape)
print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.4'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
[convolutional.SeparableConv2D.5] 4 3x3 filters on 5x5x2 input, strides=(1,1), padding='same', data_format='channels_last', depth_multiplier=2, activation='relu', use_bias=False
In [9]:
data_in_shape = (5, 5, 2)
conv = SeparableConv2D(4, (3,3), strides=(1,1),
padding='same', data_format='channels_last',
depth_multiplier=2, activation='relu', use_bias=False)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(165)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
# print('b shape:', weights[2].shape)
# print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.5'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
[convolutional.SeparableConv2D.6] 4 3x3 filters on 5x5x2 input, strides=(2,2), padding='same', data_format='channels_last', depth_multiplier=2, activation='relu', use_bias=True
In [10]:
data_in_shape = (5, 5, 2)
conv = SeparableConv2D(4, (3,3), strides=(2,2),
padding='same', data_format='channels_last',
depth_multiplier=2, activation='relu', use_bias=True)
layer_0 = Input(shape=data_in_shape)
layer_1 = conv(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)
# set weights to random (use seed for reproducibility)
weights = []
for w in model.get_weights():
np.random.seed(166)
weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('depthwise_kernel shape:', weights[0].shape)
print('depthwise_kernel:', format_decimal(weights[0].ravel().tolist()))
print('pointwise_kernel shape:', weights[1].shape)
print('pointwise_kernel:', format_decimal(weights[1].ravel().tolist()))
print('b shape:', weights[2].shape)
print('b:', format_decimal(weights[2].ravel().tolist()))
data_in = 2 * np.random.random(data_in_shape) - 1
result = model.predict(np.array([data_in]))
data_out_shape = result[0].shape
data_in_formatted = format_decimal(data_in.ravel().tolist())
data_out_formatted = format_decimal(result[0].ravel().tolist())
print('')
print('in shape:', data_in_shape)
print('in:', data_in_formatted)
print('out shape:', data_out_shape)
print('out:', data_out_formatted)
DATA['convolutional.SeparableConv2D.6'] = {
'input': {'data': data_in_formatted, 'shape': data_in_shape},
'weights': [{'data': format_decimal(w.ravel().tolist()), 'shape': w.shape} for w in weights],
'expected': {'data': data_out_formatted, 'shape': data_out_shape}
}
In [11]:
print(json.dumps(DATA))
In [ ]: