In [1]:

    

import numpy as np
import json
from keras.models import Model
from keras.layers import Input, Activation
from keras.layers.convolutional import Conv2D, SeparableConv2D, ZeroPadding2D, Cropping2D
from keras.layers.pooling import AveragePooling2D
from keras.layers.normalization import BatchNormalization
from keras import backend as K
from collections import OrderedDict


    

    




Using TensorFlow backend.
/home/leon/miniconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
  return f(*args, **kwds)

In [2]:

    

def format_decimal(arr, places=6):
    return [round(x * 10**places) / 10**places for x in arr]


    

In [3]:

    

DATA = OrderedDict()


    

In [4]:

    

random_seed = 10016
data_in_shape = (10,10,2)

layers = [
    Conv2D(4, (3,3), padding='valid', strides=(2,2), use_bias=False),
    BatchNormalization(epsilon=1e-03, axis=-1, center=True, scale=True),
    Activation('relu'),
    ZeroPadding2D(padding=((0, 1), (0, 1))),
    Cropping2D(cropping=((1, 0), (1, 0))),
    AveragePooling2D(pool_size=(1, 1), padding='valid', strides=(2, 2))
]

input_layer = Input(shape=data_in_shape)
x = layers[0](input_layer)
for layer in layers[1:-1]:
    x = layer(x)
output_layer = layers[-1](x)
model = Model(inputs=input_layer, outputs=output_layer)

np.random.seed(random_seed)
data_in = 2 * np.random.random(data_in_shape) - 1

# set weights to random (use seed for reproducibility)
weights = []
for i, w in enumerate(model.get_weights()):
    np.random.seed(random_seed + i)
    if i == 4:
        # std should be positive
        weights.append(np.random.random(w.shape))
    else:
        weights.append(2 * np.random.random(w.shape) - 1)
model.set_weights(weights)

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())

DATA['pipeline_16'] = {
    '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 [5]:

    

import os

filename = '../../test/data/pipeline/16.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 [6]:

    

print(json.dumps(DATA))


    

    




{"pipeline_16": {"input": {"data": [-0.296519, 0.572188, -0.272211, -0.907259, -0.996198, 0.61739, 0.110548, 0.580901, -0.115354, 0.574469, -0.03939, 0.348826, -0.042372, 0.132598, -0.620505, 0.302978, -0.907868, 0.723517, -0.666202, -0.655146, -0.169661, -0.721866, 0.65103, -0.379465, 0.187194, 0.910984, -0.046393, -0.717226, -0.220449, -0.812018, 0.758602, -0.425723, 0.826973, 0.41357, -0.982014, 0.308827, 0.930297, 0.552791, 0.422404, -0.45334, 0.271448, 0.563136, -0.316318, 0.641678, 0.363059, -0.440587, -0.037204, -0.119982, 0.826628, -0.369548, 0.06503, 0.559157, 0.855733, 0.156159, 0.690997, 0.38689, 0.197399, -0.520678, -0.60102, -0.519463, 0.437547, -0.963725, 0.840374, -0.961812, 0.682657, 0.683939, 0.6797, 0.823475, -0.155763, 0.407472, 0.890151, -0.275848, -0.321692, -0.669862, 0.847674, 0.467442, -0.767249, -0.762515, 0.856726, -0.318873, 0.694542, 0.714746, -0.143808, -0.049621, -0.475187, -0.353159, 0.74349, 0.257662, -0.882949, -0.242672, -0.152523, -0.014676, 0.055601, 0.544025, -0.020051, -0.797163, -0.304542, 0.614325, -0.53998, 0.375662, -0.705527, -0.520286, -0.7861, -0.567627, -0.367321, 0.327227, 0.283708, 0.457965, 0.984347, -0.337644, -0.675635, -0.201958, -0.396271, 0.353544, 0.598736, -0.595224, 0.687218, 0.23673, 0.212353, 0.278087, 0.918215, -0.460093, 0.023297, -0.691769, -0.754752, -0.38847, 0.136175, 0.73544, -0.589597, -0.838913, 0.145886, 0.840253, -0.86358, 0.11517, -0.739268, -0.728574, -0.113248, -0.218071, -0.380281, -0.124292, 0.260543, -0.949342, 0.21152, 0.003548, 0.119453, 0.589717, 0.369072, 0.470988, 0.829205, 0.815702, -0.537322, -0.762229, 0.627759, 0.996532, 0.312333, -0.70115, -0.706727, 0.248581, -0.005799, 0.076031, -0.705038, 0.652309, -0.132503, -0.653827, -0.347404, -0.962842, -0.264451, 0.624884, -0.894047, -0.42713, 0.970572, 0.377402, -0.190215, 0.065945, -0.620644, 0.938643, 0.312769, 0.309219, 0.45483, 0.912505, 0.654109, 0.086082, -0.044841, -0.662166, 0.669878, -0.421224, -0.663752, 0.737268, -0.133045, 0.045642, 0.608814, 0.486256, 0.537827, 0.787655, -0.942509, -0.339053, -0.01045, -0.764766, -0.35555, -0.729608], "shape": [10, 10, 2]}, "weights": [{"data": [-0.296519, 0.572188, -0.272211, -0.907259, -0.996198, 0.61739, 0.110548, 0.580901, -0.115354, 0.574469, -0.03939, 0.348826, -0.042372, 0.132598, -0.620505, 0.302978, -0.907868, 0.723517, -0.666202, -0.655146, -0.169661, -0.721866, 0.65103, -0.379465, 0.187194, 0.910984, -0.046393, -0.717226, -0.220449, -0.812018, 0.758602, -0.425723, 0.826973, 0.41357, -0.982014, 0.308827, 0.930297, 0.552791, 0.422404, -0.45334, 0.271448, 0.563136, -0.316318, 0.641678, 0.363059, -0.440587, -0.037204, -0.119982, 0.826628, -0.369548, 0.06503, 0.559157, 0.855733, 0.156159, 0.690997, 0.38689, 0.197399, -0.520678, -0.60102, -0.519463, 0.437547, -0.963725, 0.840374, -0.961812, 0.682657, 0.683939, 0.6797, 0.823475, -0.155763, 0.407472, 0.890151, -0.275848], "shape": [3, 3, 2, 4]}, {"data": [-0.983721, 0.040472, -0.976147, -0.023768], "shape": [4]}, {"data": [0.524698, 0.474407, 0.471901, 0.222861], "shape": [4]}, {"data": [-0.040091, -0.549528, 0.347624, 0.057999], "shape": [4]}, {"data": [0.14783, 0.198852, 0.395548, 0.513733], "shape": [4]}], "expected": {"data": [0.261207, 0.531501, 4.084157, 0.351517, 0.429899, 0.814022, 3.722062, 0.208077, 0.0, 0.393933, 3.271786, 0.231966, 0.034173, 0.340105, 0.0, 0.254722], "shape": [2, 2, 4]}}}

In [ ]: