model-optimizer/extensions/middle/EltwiseInputReshape_test.py

   1 """
   2  Copyright (c) 2018-2019 Intel Corporation
   3
   4  Licensed under the Apache License, Version 2.0 (the "License");
   5  you may not use this file except in compliance with the License.
   6  You may obtain a copy of the License at
   7
   8       http://www.apache.org/licenses/LICENSE-2.0
   9
  10  Unless required by applicable law or agreed to in writing, software
  11  distributed under the License is distributed on an "AS IS" BASIS,
  12  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  See the License for the specific language governing permissions and
  14  limitations under the License.
  15 """
  16
  17 import unittest
  18
  19 import numpy as np
  20
  21 from extensions.middle.EltwiseInputReshape import EltwiseInputReshape
  22 from mo.middle.passes.eliminate_test import build_graph
  23 from mo.middle.passes.fusing.fuse_linear_ops_test import compare_graphs
  24
  25 # The dictionary with nodes attributes used to build various graphs. A key is the name of the node and the value is the
  26 # dictionary with node attributes.
  27 nodes_attributes = {
  28     # Placeholder layers
  29     'placeholder_1': {'value': None, 'shape': None, 'type': 'Placeholder', 'kind': 'op', 'op': 'Placeholder'},
  30     'placeholder_1_data': {'value': None, 'shape': None, 'kind': 'data', 'data_type': None},
  31     'placeholder_2_data': {'value': None, 'shape': None, 'kind': 'data', 'data_type': None},
  32     'placeholder_3_data': {'value': None, 'shape': None, 'kind': 'data', 'data_type': None},
  33
  34     # Reshape layers
  35     'reshape_1': {'type': 'Reshape', 'value': None, 'kind': 'op', 'op': 'Reshape'},
  36     'reshape_1_data': {'value': None, 'shape': None, 'kind': 'data'},
  37
  38     'reshape_2': {'type': 'Reshape', 'value': None, 'kind': 'op', 'op': 'Reshape'},
  39     'reshape_2_data': {'value': None, 'shape': None, 'kind': 'data'},
  40
  41     # Fake consumes layers
  42     'consumer_1': {'type': 'Consumer', 'value': None, 'kind': 'op', 'op': 'Consumer'},
  43     'consumer_2': {'type': 'Consumer', 'value': None, 'kind': 'op', 'op': 'Consumer'},
  44     'consumer_3': {'type': 'Consumer', 'value': None, 'kind': 'op', 'op': 'Consumer'},
  45
  46     # Concat
  47     'concat': {'type': 'Concat', 'kind': 'op', 'op': 'Concat'},
  48 }
  49
  50
  51 class EltwiseInputReshapeTest(unittest.TestCase):
  52     def test1_not_constant(self):
  53         #        ,-------------->consumer3                 ,------------>consumer3
  54         #   data---(new_shape1)-->consumer1      =>    data---->Reshape-->consumer1
  55         #        `-(new_shape2)-->consumer2                 `-->Reshape-->consumer2
  56         #
  57         graph = build_graph(nodes_attributes,
  58                             [('placeholder_1_data', 'consumer_1', {'new_shape': [1, 3, 1, 1]}),
  59                              ('placeholder_1_data', 'consumer_2', {'new_shape': [1, 1, 3]}),
  60                              ('placeholder_1_data', 'consumer_3'),
  61                              ('consumer_1', 'concat'),
  62                              ('consumer_2', 'concat'),
  63                              ('consumer_3', 'concat'),
  64                              ],
  65                             {'placeholder_1_data': {'shape': np.array([1, 3])}}, nodes_with_edges_only=True)
  66
  67         graph_ref = build_graph(nodes_attributes,
  68                                 [('placeholder_1_data', 'reshape_1'),
  69                                  ('placeholder_1_data', 'reshape_2'),
  70                                  ('placeholder_1_data', 'consumer_3'),
  71                                  ('reshape_1', 'reshape_1_data'),
  72                                  ('reshape_2', 'reshape_2_data'),
  73                                  ('reshape_1_data', 'consumer_1'),
  74                                  ('reshape_2_data', 'consumer_2'),
  75                                  ('consumer_1', 'concat'),
  76                                  ('consumer_2', 'concat'),
  77                                  ('consumer_3', 'concat'),
  78                                  ],
  79                                 {'placeholder_1_data': {'shape': np.array([1, 3])},
  80                                  'reshape_1': {'dim': np.array([1, 3, 1, 1])},
  81                                  'reshape_1_data': {'shape': np.array([1, 3, 1, 1])},
  82                                  'reshape_2': {'dim': np.array([1, 1, 3])},
  83                                  'reshape_2_data': {'shape': np.array([1, 1, 3])},
  84                                  }, nodes_with_edges_only=True)
  85
  86         pattern = EltwiseInputReshape()
  87         pattern.find_and_replace_pattern(graph)
  88
  89         (flag, resp) = compare_graphs(graph, graph_ref, 'concat', check_op_attrs=True)
  90         self.assertTrue(flag, resp)
  91
  92     def test2_not_constant(self):
  93         #        ,--------------->consumer3                ,----------->consumer3
  94         #   data---(new_shape1)-->consumer1      =>    data-->Reshape-->consumer1
  95         #        `-(new_shape1)-->consumer2                         `-->consumer2
  96         #
  97         graph = build_graph(nodes_attributes,
  98                             [('placeholder_1_data', 'consumer_1', {'new_shape': [1, 3, 1, 1]}),
  99                              ('placeholder_1_data', 'consumer_2', {'new_shape': [1, 3, 1, 1]}),
 100                              ('placeholder_1_data', 'consumer_3'),
 101                              ('consumer_1', 'concat'),
 102                              ('consumer_2', 'concat'),
 103                              ('consumer_3', 'concat'),
 104                              ],
 105                             {'placeholder_1_data': {'shape': np.array([1, 3])}}, nodes_with_edges_only=True)
 106
 107         graph_ref = build_graph(nodes_attributes,
 108                                 [('placeholder_1_data', 'reshape_1'),
 109                                  ('placeholder_1_data', 'consumer_3'),
 110                                  ('reshape_1', 'reshape_1_data'),
 111                                  ('reshape_1_data', 'consumer_1'),
 112                                  ('reshape_1_data', 'consumer_2'),
 113                                  ('consumer_1', 'concat'),
 114                                  ('consumer_2', 'concat'),
 115                                  ('consumer_3', 'concat'),
 116                                  ],
 117                                 {'placeholder_1_data': {'shape': np.array([1, 3])},
 118                                  'reshape_1': {'dim': np.array([1, 3, 1, 1])},
 119                                  'reshape_1_data': {'shape': np.array([1, 3, 1, 1])},
 120                                  }, nodes_with_edges_only=True)
 121
 122         pattern = EltwiseInputReshape()
 123         pattern.find_and_replace_pattern(graph)
 124
 125         (flag, resp) = compare_graphs(graph, graph_ref, 'concat', check_op_attrs=True)
 126         self.assertTrue(flag, resp)
 127
 128     def test3_constant(self):
 129         #        ,--------------->consumer3            data-->consumer3
 130         #   data---(new_shape1)-->consumer1      =>    data-->consumer1
 131         #        `-(new_shape2)-->consumer2            data-->consumer2
 132         #
 133         graph = build_graph(nodes_attributes,
 134                             [('placeholder_1_data', 'consumer_1', {'new_shape': [1, 3, 1, 1]}),
 135                              ('placeholder_1_data', 'consumer_2', {'new_shape': [1, 1, 3]}),
 136                              ('placeholder_1_data', 'consumer_3'),
 137                              ('consumer_1', 'concat'),
 138                              ('consumer_2', 'concat'),
 139                              ('consumer_3', 'concat'),
 140                              ],
 141                             {'placeholder_1_data': {'shape': np.array([1, 3]), 'value': np.ones([1, 3])}},
 142                             nodes_with_edges_only=True)
 143
 144         graph_ref = build_graph(nodes_attributes,
 145                                 [('placeholder_1_data', 'consumer_1'),
 146                                  ('placeholder_2_data', 'consumer_2'),
 147                                  ('placeholder_3_data', 'consumer_3'),
 148                                  ('consumer_1', 'concat'),
 149                                  ('consumer_2', 'concat'),
 150                                  ('consumer_3', 'concat'),
 151                                  ],
 152                                 {'placeholder_1_data': {'shape': np.array([1, 3, 1, 1]), 'value': np.ones([1, 3, 1, 1])},
 153                                  'placeholder_2_data': {'shape': np.array([1, 1, 3]), 'value': np.ones([1, 1, 3])},
 154                                  'placeholder_3_data': {'shape': np.array([1, 3]), 'value': np.ones([1, 3])},
 155                                  }, nodes_with_edges_only=True)
 156
 157         pattern = EltwiseInputReshape()
 158         pattern.find_and_replace_pattern(graph)
 159
 160         (flag, resp) = compare_graphs(graph, graph_ref, 'concat', check_op_attrs=True)
 161         self.assertTrue(flag, resp)
 162
 163     def test4_constant(self):
 164         #        ,--------------->consumer3                 ,-->consumer3
 165         #   data---(new_shape1)-->consumer1      =>    data-->consumer1
 166         #        `-(new_shape2)-->consumer2                 `->consumer2
 167         #
 168         graph = build_graph(nodes_attributes,
 169                             [('placeholder_1_data', 'consumer_1', {'new_shape': [3, 1, 1]}),
 170                              ('placeholder_1_data', 'consumer_2', {'new_shape': [3, 1, 1]}),
 171                              ('placeholder_1_data', 'consumer_3', {'new_shape': [3, 1, 1]}),
 172                              ('consumer_1', 'concat'),
 173                              ('consumer_2', 'concat'),
 174                              ('consumer_3', 'concat'),
 175                              ],
 176                             {'placeholder_1_data': {'shape': np.array([1, 3]), 'value': np.ones([1, 3])}},
 177                             nodes_with_edges_only=True)
 178
 179         graph_ref = build_graph(nodes_attributes,
 180                                 [('placeholder_1_data', 'consumer_1'),
 181                                  ('placeholder_1_data', 'consumer_2'),
 182                                  ('placeholder_1_data', 'consumer_3'),
 183                                  ('consumer_1', 'concat'),
 184                                  ('consumer_2', 'concat'),
 185                                  ('consumer_3', 'concat'),
 186                                  ],
 187                                 {'placeholder_1_data': {'shape': np.array([3, 1, 1]), 'value': np.ones([3, 1, 1])}
 188                                  }, nodes_with_edges_only=True)
 189
 190         pattern = EltwiseInputReshape()
 191         pattern.find_and_replace_pattern(graph)
 192
 193         (flag, resp) = compare_graphs(graph, graph_ref, 'concat', check_op_attrs=True)
 194         self.assertTrue(flag, resp)
 195
 196     def test5_not_constant(self):
 197         #        ,--------------->consumer3                ,->consumer3
 198         #   data---(new_shape1)-->consumer1      =>    data----->consumer1
 199         #        `-(new_shape1)-->consumer2                `-->consumer2
 200         #
 201         graph = build_graph(nodes_attributes,
 202                             [('placeholder_1_data', 'consumer_1', {'new_shape': [1, 3]}),
 203                              ('placeholder_1_data', 'consumer_2', {'new_shape': [1, 3]}),
 204                              ('placeholder_1_data', 'consumer_3'),
 205                              ('consumer_1', 'concat'),
 206                              ('consumer_2', 'concat'),
 207                              ('consumer_3', 'concat'),
 208                              ],
 209                             {'placeholder_1_data': {'shape': np.array([1, 3])}}, nodes_with_edges_only=True)
 210
 211         graph_ref = build_graph(nodes_attributes,
 212                             [('placeholder_1_data', 'consumer_1', {'new_shape': [1, 3]}),
 213                              ('placeholder_1_data', 'consumer_2', {'new_shape': [1, 3]}),
 214                              ('placeholder_1_data', 'consumer_3'),
 215                              ('consumer_1', 'concat'),
 216                              ('consumer_2', 'concat'),
 217                              ('consumer_3', 'concat'),
 218                              ],
 219                             {'placeholder_1_data': {'shape': np.array([1, 3])}}, nodes_with_edges_only=True)
 220
 221         pattern = EltwiseInputReshape()
 222         pattern.find_and_replace_pattern(graph)
 223
 224         (flag, resp) = compare_graphs(graph, graph_ref, 'concat', check_op_attrs=True)
 225         self.assertTrue(flag, resp)