

In [1]:

    
import numpy as np
from keras.models import Model
from keras.layers import Input
from keras.layers.normalization import BatchNormalization
from keras import backend as K
import json
from collections import OrderedDict









    



Using TensorFlow backend.



In [2]:

    
def format_decimal(arr, places=8):
    return [round(x * 10**places) / 10**places for x in arr]



In [3]:

    
DATA = OrderedDict()

BatchNormalization

[normalization.BatchNormalization.0] epsilon=1e-05, axis=-1, center=True, scale=True



In [4]:

    
data_in_shape = (4, 3)
norm = BatchNormalization(epsilon=1e-05, axis=-1, center=True, scale=True)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1000 + i)
    if i == 3:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('gamma shape:', weights[0].shape)
print('gamma:', format_decimal(weights[0].ravel().tolist()))
print('beta shape:', weights[1].shape)
print('beta:', format_decimal(weights[1].ravel().tolist()))
print('moving_mean shape:', weights[2].shape)
print('moving_mean:', format_decimal(weights[2].ravel().tolist()))
print('moving_variance shape:', weights[3].shape)
print('moving_variance:', format_decimal(weights[3].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['normalization.BatchNormalization.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}
}









    



gamma shape: (3,)
gamma: [0.30717917, -0.76998611, 0.90056573]
beta shape: (3,)
beta: [-0.38753564, -0.46987287, -0.60787987]
moving_mean shape: (3,)
moving_mean: [-0.74202342, -0.07768763, -0.16769214]
moving_variance shape: (3,)
moving_variance: [0.5978058, 0.43593379, 0.01686999]

in shape: (4, 3)
in: [0.19337492, 0.7899562, 0.06925513, -0.98808939, 0.80435901, 0.50903924, -0.65579245, 0.46005824, -0.25374991, -0.63537441, -0.10931755, -0.42626602]
out shape: (4, 3)
out: [-0.01591084, -1.48170662, 1.03452778, -0.48529527, -1.49850297, 4.08290577, -0.35327691, -1.09698439, -1.20439208, -0.34516501, -0.4329865, -2.40019298]

[normalization.BatchNormalization.1] epsilon=1e-02, axis=-1, center=True, scale=True



In [5]:

    
data_in_shape = (4, 3)
norm = BatchNormalization(epsilon=1e-02, axis=-1, center=True, scale=True)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1010 + i)
    if i == 3:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('gamma shape:', weights[0].shape)
print('gamma:', format_decimal(weights[0].ravel().tolist()))
print('beta shape:', weights[1].shape)
print('beta:', format_decimal(weights[1].ravel().tolist()))
print('moving_mean shape:', weights[2].shape)
print('moving_mean:', format_decimal(weights[2].ravel().tolist()))
print('moving_variance shape:', weights[3].shape)
print('moving_variance:', format_decimal(weights[3].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['normalization.BatchNormalization.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}
}









    



gamma shape: (3,)
gamma: [-0.21148703, -0.64881506, -0.85458828]
beta shape: (3,)
beta: [0.31136173, -0.22851883, 0.25302429]
moving_mean shape: (3,)
moving_mean: [-0.94654129, 0.58559264, -0.49527049]
moving_variance shape: (3,)
moving_variance: [0.55703859, 0.05517086, 0.26398732]

in shape: (4, 3)
in: [0.71880835, -0.1596278, -0.95536448, -0.94580233, -0.43370689, -0.55992129, 0.41852567, 0.03566775, 0.0999246, 0.53704832, 0.9145912, -0.50412891]
out shape: (4, 3)
out: [-0.15635495, 1.66547668, 1.00419426, 0.31115419, 2.3620553, 0.358576, -0.07201998, 1.16912842, -0.71871787, -0.1053073, -1.06467664, 0.26748699]

[normalization.BatchNormalization.2] epsilon=1e-05, axis=1, center=True, scale=True



In [6]:

    
data_in_shape = (4, 3, 2)
norm = BatchNormalization(epsilon=1e-05, axis=1, center=True, scale=True)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1020 + i)
    if i == 3:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('gamma shape:', weights[0].shape)
print('gamma:', format_decimal(weights[0].ravel().tolist()))
print('beta shape:', weights[1].shape)
print('beta:', format_decimal(weights[1].ravel().tolist()))
print('moving_mean shape:', weights[2].shape)
print('moving_mean:', format_decimal(weights[2].ravel().tolist()))
print('moving_variance shape:', weights[3].shape)
print('moving_variance:', format_decimal(weights[3].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['normalization.BatchNormalization.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}
}









    



gamma shape: (4,)
gamma: [0.51790609, -0.66653736, 0.37866461, -0.38006242]
beta shape: (4,)
beta: [0.40055717, 0.74387118, 0.43713362, -0.55480014]
moving_mean shape: (4,)
moving_mean: [-0.72639303, 0.96140456, -0.35265082, -0.61683149]
moving_variance shape: (4,)
moving_variance: [0.91406979, 0.63007121, 0.50893322, 0.05289729]

in shape: (4, 3, 2)
in: [-0.92002272, 0.46666412, 0.33270192, 0.69369639, -0.70539125, 0.38574141, 0.2966058, 0.80575415, -0.87667886, -0.14090813, -0.81737496, 0.22243293, -0.40642584, 0.14401339, 0.56671486, 0.21548293, -0.18429046, 0.90187812, 0.31491955, -0.56054439, 0.98960233, 0.26554054, 0.75526676, 0.66461658]
out shape: (4, 3, 2)
out: [0.29566789, 1.04683638, 0.97426903, 1.16981983, 0.41193381, 1.00300062, 1.30210531, 0.87457144, 2.28731728, 1.66948748, 2.23751926, 1.36438859, 0.40859056, 0.70075637, 0.92512071, 0.73869145, 0.52649707, 1.10302091, -2.09436178, -0.64780515, -3.20916128, -2.01277113, -2.8219614, -2.67217731]

[normalization.BatchNormalization.3] epsilon=1e-05, axis=2, center=True, scale=True



In [7]:

    
data_in_shape = (4, 3, 2)
norm = BatchNormalization(epsilon=1e-05, axis=2, center=True, scale=True)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1030 + i)
    if i == 3:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('gamma shape:', weights[0].shape)
print('gamma:', format_decimal(weights[0].ravel().tolist()))
print('beta shape:', weights[1].shape)
print('beta:', format_decimal(weights[1].ravel().tolist()))
print('moving_mean shape:', weights[2].shape)
print('moving_mean:', format_decimal(weights[2].ravel().tolist()))
print('moving_variance shape:', weights[3].shape)
print('moving_variance:', format_decimal(weights[3].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['normalization.BatchNormalization.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}
}









    



gamma shape: (3,)
gamma: [-0.08987256, -0.21030534, -0.5544436]
beta shape: (3,)
beta: [-0.99796406, -0.48734278, -0.35036232]
moving_mean shape: (3,)
moving_mean: [-0.14543666, 0.11871984, -0.36856312]
moving_variance shape: (3,)
moving_variance: [0.48693282, 0.58735461, 0.51704096]

in shape: (4, 3, 2)
in: [0.17723912, -0.81800371, -0.95525628, -0.62524101, 0.9361622, 0.47178858, -0.25140505, 0.6715247, -0.81975913, 0.46866456, 0.49221768, 0.01473235, -0.4159851, 0.35468316, -0.44554467, -0.40155922, 0.68942038, 0.30257063, -0.24790617, -0.87167286, -0.15334632, -0.94018827, 0.92413894, 0.55480878]
out shape: (4, 3, 2)
out: [-1.03952205, -0.91134298, -0.19263539, -0.28319412, -1.35638976, -0.99832773, -0.98431623, -1.10318196, -0.22981687, -0.58337033, -1.01407993, -0.64590788, -0.96311969, -1.06237543, -0.33250421, -0.34457415, -1.16613591, -0.86784983, -0.98476684, -0.90443087, -0.41268572, -0.19677016, -1.34711909, -1.06234169]

[normalization.BatchNormalization.4] epsilon=1e-05, axis=3, center=True, scale=False



In [8]:

    
data_in_shape = (4, 3, 2)
norm = BatchNormalization(epsilon=1e-05, axis=3, center=True, scale=False)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1040 + i)
    if i == 2:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('beta shape:', weights[0].shape)
print('beta:', format_decimal(weights[0].ravel().tolist()))
print('moving_mean shape:', weights[1].shape)
print('moving_mean:', format_decimal(weights[1].ravel().tolist()))
print('moving_variance shape:', weights[2].shape)
print('moving_variance:', 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['normalization.BatchNormalization.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}
}









    



beta shape: (2,)
beta: [0.6605443, -0.04771633]
moving_mean shape: (2,)
moving_mean: [-0.95665551, 0.1278139]
moving_variance shape: (2,)
moving_variance: [0.32031337, 0.14493475]

in shape: (4, 3, 2)
in: [0.64108404, -0.67698454, 0.90018603, -0.32716791, -0.69711497, 0.3571678, -0.53864991, 0.94118592, 0.90909098, -0.35280573, -0.50526634, -0.61466515, -0.66641796, 0.25639848, -0.50969593, -0.75285121, 0.34839266, 0.46978101, -0.1664059, -0.08380992, 0.63740243, -0.02691944, -0.09122654, -0.79602303]
out shape: (4, 3, 2)
out: [3.48354912, -2.16162324, 3.94134998, -1.24278474, 1.11911988, 0.55471134, 1.39910769, 2.08871031, 3.95708394, -1.31012583, 1.45809221, -1.99793351, 1.17335761, 0.29002765, 1.45026565, -2.36089683, 2.96640038, 0.8505044, 2.0568161, -0.60357362, 3.47704434, -0.45414343, 2.18964863, -2.47429323]

[normalization.BatchNormalization.5] epsilon=1e-05, axis=-1, center=False, scale=True



In [9]:

    
data_in_shape = (4, 3, 2)
norm = BatchNormalization(epsilon=1e-05, axis=-1, center=False, scale=True)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1050 + i)
    if i == 2:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('gamma shape:', weights[0].shape)
print('gamma:', format_decimal(weights[0].ravel().tolist()))
print('moving_mean shape:', weights[1].shape)
print('moving_mean:', format_decimal(weights[1].ravel().tolist()))
print('moving_variance shape:', weights[2].shape)
print('moving_variance:', 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['normalization.BatchNormalization.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}
}









    



gamma shape: (2,)
gamma: [0.8332942, -0.73282841]
moving_mean shape: (2,)
moving_mean: [0.01371764, -0.84259866]
moving_variance shape: (2,)
moving_variance: [0.66551526, 0.38091611]

in shape: (4, 3, 2)
in: [0.29214552, -0.66840164, 0.52167053, 0.49733931, 0.88648322, 0.55305475, 0.0485971, -0.33068467, 0.11495103, 0.26902905, -0.43824223, 0.98597112, 0.8185068, -0.38642272, -0.02336415, 0.39224186, 0.95561884, 0.66953993, -0.85531119, 0.83512858, 0.60595255, 0.04343527, -0.48985708, 0.45544707]
out shape: (4, 3, 2)
out: [0.28439951, -0.20683438, 0.51884729, -1.59098697, 0.89148432, -1.65714121, 0.03562754, -0.60782552, 0.1034046, -1.31990075, -0.46165335, -2.17116809, 0.82204998, -0.54164445, -0.03787711, -1.46619856, 0.96210277, -1.79545081, -0.88766748, -1.992064, 0.60493696, -1.0520401, -0.51437521, -1.54124582]

[normalization.BatchNormalization.6] epsilon=0.001, axis=-1, center=False, scale=False



In [10]:

    
data_in_shape = (4, 3, 2)
norm = BatchNormalization(epsilon=0.001, axis=-1, center=False, scale=False)

layer_0 = Input(shape=data_in_shape)
layer_1 = norm(layer_0)
model = Model(inputs=layer_0, outputs=layer_1)

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(1060 + i)
    if i == 1:
        # variance should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)
print('moving_mean shape:', weights[0].shape)
print('moving_mean:', format_decimal(weights[0].ravel().tolist()))
print('moving_variance shape:', weights[1].shape)
print('moving_variance:', format_decimal(weights[1].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['normalization.BatchNormalization.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}
}









    



moving_mean shape: (2,)
moving_mean: [0.12870543, -0.95606727]
moving_variance shape: (2,)
moving_variance: [0.41595926, 0.00524988]

in shape: (4, 3, 2)
in: [-0.09731108, -0.49690708, 0.92191878, -0.77721125, -0.94202527, -0.89542274, 0.78510398, 0.47090425, -0.60424113, 0.33366004, 0.16897797, 0.86414392, -0.97380934, 0.98724401, 0.72680442, 0.74601726, 0.80377242, -0.69024304, -0.78895346, -0.89545572, -0.37614091, -0.69877677, 0.11618973, 0.78171578]
out shape: (4, 3, 2)
out: [-0.35002038, 5.80802298, 1.22840953, 2.26239109, -1.65818667, 0.76710606, 1.01653147, 18.0500927, -1.13507748, 16.3140583, 0.06236805, 23.02427101, -1.70740914, 24.5813942, 0.92624581, 21.53005981, 1.04544234, 3.36247158, -1.42113221, 0.7666893, -0.78183013, 3.25452614, -0.01938242, 21.98161888]

export for Keras.js tests



In [11]:

    
import os

filename = '../../../test/data/layers/normalization/BatchNormalization.json'
if not os.path.exists(os.path.dirname(filename)):
    os.makedirs(os.path.dirname(filename))
with open(filename, 'w') as f:
    json.dump(DATA, f)



In [12]:

    
print(json.dumps(DATA))









    



{"normalization.BatchNormalization.0": {"input": {"data": [0.19337492, 0.7899562, 0.06925513, -0.98808939, 0.80435901, 0.50903924, -0.65579245, 0.46005824, -0.25374991, -0.63537441, -0.10931755, -0.42626602], "shape": [4, 3]}, "weights": [{"data": [0.30717917, -0.76998611, 0.90056573], "shape": [3]}, {"data": [-0.38753564, -0.46987287, -0.60787987], "shape": [3]}, {"data": [-0.74202342, -0.07768763, -0.16769214], "shape": [3]}, {"data": [0.5978058, 0.43593379, 0.01686999], "shape": [3]}], "expected": {"data": [-0.01591084, -1.48170662, 1.03452778, -0.48529527, -1.49850297, 4.08290577, -0.35327691, -1.09698439, -1.20439208, -0.34516501, -0.4329865, -2.40019298], "shape": [4, 3]}}, "normalization.BatchNormalization.1": {"input": {"data": [0.71880835, -0.1596278, -0.95536448, -0.94580233, -0.43370689, -0.55992129, 0.41852567, 0.03566775, 0.0999246, 0.53704832, 0.9145912, -0.50412891], "shape": [4, 3]}, "weights": [{"data": [-0.21148703, -0.64881506, -0.85458828], "shape": [3]}, {"data": [0.31136173, -0.22851883, 0.25302429], "shape": [3]}, {"data": [-0.94654129, 0.58559264, -0.49527049], "shape": [3]}, {"data": [0.55703859, 0.05517086, 0.26398732], "shape": [3]}], "expected": {"data": [-0.15635495, 1.66547668, 1.00419426, 0.31115419, 2.3620553, 0.358576, -0.07201998, 1.16912842, -0.71871787, -0.1053073, -1.06467664, 0.26748699], "shape": [4, 3]}}, "normalization.BatchNormalization.2": {"input": {"data": [-0.92002272, 0.46666412, 0.33270192, 0.69369639, -0.70539125, 0.38574141, 0.2966058, 0.80575415, -0.87667886, -0.14090813, -0.81737496, 0.22243293, -0.40642584, 0.14401339, 0.56671486, 0.21548293, -0.18429046, 0.90187812, 0.31491955, -0.56054439, 0.98960233, 0.26554054, 0.75526676, 0.66461658], "shape": [4, 3, 2]}, "weights": [{"data": [0.51790609, -0.66653736, 0.37866461, -0.38006242], "shape": [4]}, {"data": [0.40055717, 0.74387118, 0.43713362, -0.55480014], "shape": [4]}, {"data": [-0.72639303, 0.96140456, -0.35265082, -0.61683149], "shape": [4]}, {"data": [0.91406979, 0.63007121, 0.50893322, 0.05289729], "shape": [4]}], "expected": {"data": [0.29566789, 1.04683638, 0.97426903, 1.16981983, 0.41193381, 1.00300062, 1.30210531, 0.87457144, 2.28731728, 1.66948748, 2.23751926, 1.36438859, 0.40859056, 0.70075637, 0.92512071, 0.73869145, 0.52649707, 1.10302091, -2.09436178, -0.64780515, -3.20916128, -2.01277113, -2.8219614, -2.67217731], "shape": [4, 3, 2]}}, "normalization.BatchNormalization.3": {"input": {"data": [0.17723912, -0.81800371, -0.95525628, -0.62524101, 0.9361622, 0.47178858, -0.25140505, 0.6715247, -0.81975913, 0.46866456, 0.49221768, 0.01473235, -0.4159851, 0.35468316, -0.44554467, -0.40155922, 0.68942038, 0.30257063, -0.24790617, -0.87167286, -0.15334632, -0.94018827, 0.92413894, 0.55480878], "shape": [4, 3, 2]}, "weights": [{"data": [-0.08987256, -0.21030534, -0.5544436], "shape": [3]}, {"data": [-0.99796406, -0.48734278, -0.35036232], "shape": [3]}, {"data": [-0.14543666, 0.11871984, -0.36856312], "shape": [3]}, {"data": [0.48693282, 0.58735461, 0.51704096], "shape": [3]}], "expected": {"data": [-1.03952205, -0.91134298, -0.19263539, -0.28319412, -1.35638976, -0.99832773, -0.98431623, -1.10318196, -0.22981687, -0.58337033, -1.01407993, -0.64590788, -0.96311969, -1.06237543, -0.33250421, -0.34457415, -1.16613591, -0.86784983, -0.98476684, -0.90443087, -0.41268572, -0.19677016, -1.34711909, -1.06234169], "shape": [4, 3, 2]}}, "normalization.BatchNormalization.4": {"input": {"data": [0.64108404, -0.67698454, 0.90018603, -0.32716791, -0.69711497, 0.3571678, -0.53864991, 0.94118592, 0.90909098, -0.35280573, -0.50526634, -0.61466515, -0.66641796, 0.25639848, -0.50969593, -0.75285121, 0.34839266, 0.46978101, -0.1664059, -0.08380992, 0.63740243, -0.02691944, -0.09122654, -0.79602303], "shape": [4, 3, 2]}, "weights": [{"data": [0.6605443, -0.04771633], "shape": [2]}, {"data": [-0.95665551, 0.1278139], "shape": [2]}, {"data": [0.32031337, 0.14493475], "shape": [2]}], "expected": {"data": [3.48354912, -2.16162324, 3.94134998, -1.24278474, 1.11911988, 0.55471134, 1.39910769, 2.08871031, 3.95708394, -1.31012583, 1.45809221, -1.99793351, 1.17335761, 0.29002765, 1.45026565, -2.36089683, 2.96640038, 0.8505044, 2.0568161, -0.60357362, 3.47704434, -0.45414343, 2.18964863, -2.47429323], "shape": [4, 3, 2]}}, "normalization.BatchNormalization.5": {"input": {"data": [0.29214552, -0.66840164, 0.52167053, 0.49733931, 0.88648322, 0.55305475, 0.0485971, -0.33068467, 0.11495103, 0.26902905, -0.43824223, 0.98597112, 0.8185068, -0.38642272, -0.02336415, 0.39224186, 0.95561884, 0.66953993, -0.85531119, 0.83512858, 0.60595255, 0.04343527, -0.48985708, 0.45544707], "shape": [4, 3, 2]}, "weights": [{"data": [0.8332942, -0.73282841], "shape": [2]}, {"data": [0.01371764, -0.84259866], "shape": [2]}, {"data": [0.66551526, 0.38091611], "shape": [2]}], "expected": {"data": [0.28439951, -0.20683438, 0.51884729, -1.59098697, 0.89148432, -1.65714121, 0.03562754, -0.60782552, 0.1034046, -1.31990075, -0.46165335, -2.17116809, 0.82204998, -0.54164445, -0.03787711, -1.46619856, 0.96210277, -1.79545081, -0.88766748, -1.992064, 0.60493696, -1.0520401, -0.51437521, -1.54124582], "shape": [4, 3, 2]}}, "normalization.BatchNormalization.6": {"input": {"data": [-0.09731108, -0.49690708, 0.92191878, -0.77721125, -0.94202527, -0.89542274, 0.78510398, 0.47090425, -0.60424113, 0.33366004, 0.16897797, 0.86414392, -0.97380934, 0.98724401, 0.72680442, 0.74601726, 0.80377242, -0.69024304, -0.78895346, -0.89545572, -0.37614091, -0.69877677, 0.11618973, 0.78171578], "shape": [4, 3, 2]}, "weights": [{"data": [0.12870543, -0.95606727], "shape": [2]}, {"data": [0.41595926, 0.00524988], "shape": [2]}], "expected": {"data": [-0.35002038, 5.80802298, 1.22840953, 2.26239109, -1.65818667, 0.76710606, 1.01653147, 18.0500927, -1.13507748, 16.3140583, 0.06236805, 23.02427101, -1.70740914, 24.5813942, 0.92624581, 21.53005981, 1.04544234, 3.36247158, -1.42113221, 0.7666893, -0.78183013, 3.25452614, -0.01938242, 21.98161888], "shape": [4, 3, 2]}}}



In [ ]: