Gitiles

review.mlplatform.org / ml / ethos-u / ml-embedded-evaluation-kit / f32a86a6969508d7a156decbed0bfc9466ad92fa / . / model_conditioning_examples / training_utils.py

Richard Burton | f32a86a | 2022-11-15 11:46:11 +0000 | [diff] [blame^] | 1 | # SPDX-FileCopyrightText: Copyright 2021 Arm Limited and/or its affiliates <open-source-office@arm.com> |

alexander | 3c79893 | 2021-03-26 21:42:19 +0000 | [diff] [blame] | 2 | # SPDX-License-Identifier: Apache-2.0 |

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 | Utility functions related to data and models that are common to all the model conditioning examples. | ||||

17 | """ | ||||

18 | import tensorflow as tf | ||||

19 | import numpy as np | ||||

20 | |||||

21 | |||||

22 | def get_data(): | ||||

23 | """Downloads and returns the pre-processed data and labels for training and testing. | ||||

24 | |||||

25 | Returns: | ||||

26 | Tuple of: (train data, train labels, test data, test labels) | ||||

27 | """ | ||||

28 | |||||

29 | # To save time we use the MNIST dataset for this example. | ||||

30 | (x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data() | ||||

31 | |||||

32 | # Convolution operations require data to have 4 dimensions. | ||||

33 | # We divide by 255 to help training and cast to float32 for TensorFlow. | ||||

34 | x_train = (x_train[..., np.newaxis] / 255.0).astype(np.float32) | ||||

35 | x_test = (x_test[..., np.newaxis] / 255.0).astype(np.float32) | ||||

36 | |||||

37 | return x_train, y_train, x_test, y_test | ||||

38 | |||||

39 | |||||

40 | def create_model(): | ||||

41 | """Create and returns a simple Keras model for training MNIST. | ||||

42 | |||||

43 | We will use a simple convolutional neural network for this example, | ||||

44 | but the model optimization methods employed should be compatible with a | ||||

45 | wide variety of CNN architectures such as Mobilenet and Inception etc. | ||||

46 | |||||

47 | Returns: | ||||

48 | Uncompiled Keras model. | ||||

49 | """ | ||||

50 | |||||

51 | keras_model = tf.keras.models.Sequential([ | ||||

52 | tf.keras.layers.Conv2D(32, 3, padding='same', input_shape=(28, 28, 1), activation=tf.nn.relu), | ||||

53 | tf.keras.layers.Conv2D(32, 3, padding='same', activation=tf.nn.relu), | ||||

54 | tf.keras.layers.MaxPool2D(), | ||||

55 | tf.keras.layers.Conv2D(32, 3, padding='same', activation=tf.nn.relu), | ||||

56 | tf.keras.layers.MaxPool2D(), | ||||

57 | tf.keras.layers.Flatten(), | ||||

58 | tf.keras.layers.Dense(units=10, activation=tf.nn.softmax) | ||||

59 | ]) | ||||

60 | |||||

61 | return keras_model |