DVC라는 것과 관련해 프로젝트를 하나 진행 시작했는데, 딥러닝 모델들을 많이 만들어야 한다. 그래서 케라스 예제의 모델들을 가져와서 테스트 해보고 코드를 정리해서 올리기로 했다.
그래서 코드를 tf.keras로 구현 & 클래스로 정리 & 잘 돌아가는지 확인, 이 세가지를 하기 위해 도커로 코드를 실행하려고 계획했다. 그 과정을 아주 간단하게 정리하려고 한다.
순서는 아래와 같이 정리된다.
1. Dockerfile을 만든다
2. model.py를 만든다
3. Dockerfile을 빌드해서 이미지를 만든다
4. 이미지를 run하여 model.py를 실행한다
1. Dockerfile을 만든다
나는 tensorflow:2.5.0 베이스로 Dockerfile을 만들었고, requirements.txt에 opencv-python 등 간단한 것을 적어두었다.
model.py라는 파일을 나중에 테스트 해볼 예정이기 때문에 CMD, ENTRYPOINT를 설정해둔다.
FROM tensorflow/tensorflow:2.5.0
WORKDIR /usr/src
COPY ./requirements.txt .
RUN pip install --no-cache-dir --upgrade pip && \
pip install -r requirements.txt
COPY . .
CMD ["model.py"]
ENTRYPOINT ["python3"]
2. model.py를 만든다
특별한 것 없이 그냥 딥러닝 모델을 summary로 확인해보는 코드
import tensorflow as tf
import tensorflow.keras.layers as layers
class ClassificationModel:
def __init__(self, image_size, num_classes):
self.input_shape = image_size + (3,)
self.num_classes = num_classes
def build_model(self):
def data_augmentation(inputs):
data_augmentation = tf.keras.Sequential(
[
layers.experimental.preprocessing.RandomFlip("horizontal"),
layers.experimental.preprocessing.RandomRotation(0.1),
]
)
return data_augmentation(inputs)
inputs = tf.keras.Input(self.input_shape)
x = data_augmentation(inputs)
x = layers.experimental.preprocessing.Rescaling(1.0 / 255)(x)
x = layers.Conv2D(32, 3, strides=2, padding="same")(x)
x = layers.BatchNormalization()(x)
x = layers.Activation("relu")(x)
x = layers.Conv2D(64, 3, padding="same")(x)
x = layers.BatchNormalization()(x)
x = layers.Activation("relu")(x)
previous_block_activation = x # Set aside residual
for size in [128, 256, 512, 728]:
x = layers.Activation("relu")(x)
x = layers.SeparableConv2D(size, 3, padding="same")(x)
x = layers.BatchNormalization()(x)
x = layers.Activation("relu")(x)
x = layers.SeparableConv2D(size, 3, padding="same")(x)
x = layers.BatchNormalization()(x)
x = layers.MaxPooling2D(3, strides=2, padding="same")(x)
# Project residual
residual = layers.Conv2D(size, 1, strides=2, padding="same")(
previous_block_activation
)
x = layers.add([x, residual]) # Add back residual
previous_block_activation = x # Set aside next residual
x = layers.SeparableConv2D(1024, 3, padding="same")(x)
x = layers.BatchNormalization()(x)
x = layers.Activation("relu")(x)
x = layers.GlobalAveragePooling2D()(x)
if self.num_classes == 2:
activation = "sigmoid"
units = 1
else:
activation = "softmax"
units = self.num_classes
x = layers.Dropout(0.5)(x)
outputs = layers.Dense(units, activation=activation)(x)
return tf.keras.Model(inputs, outputs)
if __name__ == "__main__":
image_size = (180, 180)
num_classes = 2
model = ClassificationModel(image_size=image_size, num_classes=num_classes)
model = model.build_model()
print(model.summary())
3. Dockerfile을 빌드해서 이미지를 만든다
image_name:tag와 함께 Dockerfile을 빌드한다.
$ docker build -t keras-model:0.1 .
4. 이미지를 run하여 model.py를 실행한다
도커 이미지로 model.py를 실행해본다. 실행 결과까지 아래에서 보여준다.
$ docker run -it keras-model:0.1 model.py
2021-05-30 12:49:22.007718: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 180, 180, 3) 0
__________________________________________________________________________________________________
sequential (Sequential) (None, 180, 180, 3) 0 input_1[0][0]
__________________________________________________________________________________________________
rescaling (Rescaling) (None, 180, 180, 3) 0 sequential[0][0]
__________________________________________________________________________________________________
conv2d (Conv2D) (None, 90, 90, 32) 896 rescaling[0][0]
__________________________________________________________________________________________________
batch_normalization (BatchNorma (None, 90, 90, 32) 128 conv2d[0][0]
__________________________________________________________________________________________________
activation (Activation) (None, 90, 90, 32) 0 batch_normalization[0][0]
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 90, 90, 64) 18496 activation[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, 90, 90, 64) 256 conv2d_1[0][0]
__________________________________________________________________________________________________
activation_1 (Activation) (None, 90, 90, 64) 0 batch_normalization_1[0][0]
__________________________________________________________________________________________________
activation_2 (Activation) (None, 90, 90, 64) 0 activation_1[0][0]
__________________________________________________________________________________________________
separable_conv2d (SeparableConv (None, 90, 90, 128) 8896 activation_2[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, 90, 90, 128) 512 separable_conv2d[0][0]
__________________________________________________________________________________________________
activation_3 (Activation) (None, 90, 90, 128) 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
separable_conv2d_1 (SeparableCo (None, 90, 90, 128) 17664 activation_3[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, 90, 90, 128) 512 separable_conv2d_1[0][0]
__________________________________________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 45, 45, 128) 0 batch_normalization_3[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 45, 45, 128) 8320 activation_1[0][0]
__________________________________________________________________________________________________
add (Add) (None, 45, 45, 128) 0 max_pooling2d[0][0]
conv2d_2[0][0]
__________________________________________________________________________________________________
activation_4 (Activation) (None, 45, 45, 128) 0 add[0][0]
__________________________________________________________________________________________________
separable_conv2d_2 (SeparableCo (None, 45, 45, 256) 34176 activation_4[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 45, 45, 256) 1024 separable_conv2d_2[0][0]
__________________________________________________________________________________________________
activation_5 (Activation) (None, 45, 45, 256) 0 batch_normalization_4[0][0]
__________________________________________________________________________________________________
separable_conv2d_3 (SeparableCo (None, 45, 45, 256) 68096 activation_5[0][0]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, 45, 45, 256) 1024 separable_conv2d_3[0][0]
__________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 23, 23, 256) 0 batch_normalization_5[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 23, 23, 256) 33024 add[0][0]
__________________________________________________________________________________________________
add_1 (Add) (None, 23, 23, 256) 0 max_pooling2d_1[0][0]
conv2d_3[0][0]
__________________________________________________________________________________________________
activation_6 (Activation) (None, 23, 23, 256) 0 add_1[0][0]
__________________________________________________________________________________________________
separable_conv2d_4 (SeparableCo (None, 23, 23, 512) 133888 activation_6[0][0]
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None, 23, 23, 512) 2048 separable_conv2d_4[0][0]
__________________________________________________________________________________________________
activation_7 (Activation) (None, 23, 23, 512) 0 batch_normalization_6[0][0]
__________________________________________________________________________________________________
separable_conv2d_5 (SeparableCo (None, 23, 23, 512) 267264 activation_7[0][0]
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None, 23, 23, 512) 2048 separable_conv2d_5[0][0]
__________________________________________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None, 12, 12, 512) 0 batch_normalization_7[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 12, 12, 512) 131584 add_1[0][0]
__________________________________________________________________________________________________
add_2 (Add) (None, 12, 12, 512) 0 max_pooling2d_2[0][0]
conv2d_4[0][0]
__________________________________________________________________________________________________
activation_8 (Activation) (None, 12, 12, 512) 0 add_2[0][0]
__________________________________________________________________________________________________
separable_conv2d_6 (SeparableCo (None, 12, 12, 728) 378072 activation_8[0][0]
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None, 12, 12, 728) 2912 separable_conv2d_6[0][0]
__________________________________________________________________________________________________
activation_9 (Activation) (None, 12, 12, 728) 0 batch_normalization_8[0][0]
__________________________________________________________________________________________________
separable_conv2d_7 (SeparableCo (None, 12, 12, 728) 537264 activation_9[0][0]
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None, 12, 12, 728) 2912 separable_conv2d_7[0][0]
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None, 6, 6, 728) 0 batch_normalization_9[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 6, 6, 728) 373464 add_2[0][0]
__________________________________________________________________________________________________
add_3 (Add) (None, 6, 6, 728) 0 max_pooling2d_3[0][0]
conv2d_5[0][0]
__________________________________________________________________________________________________
separable_conv2d_8 (SeparableCo (None, 6, 6, 1024) 753048 add_3[0][0]
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None, 6, 6, 1024) 4096 separable_conv2d_8[0][0]
__________________________________________________________________________________________________
activation_10 (Activation) (None, 6, 6, 1024) 0 batch_normalization_10[0][0]
__________________________________________________________________________________________________
global_average_pooling2d (Globa (None, 1024) 0 activation_10[0][0]
__________________________________________________________________________________________________
dropout (Dropout) (None, 1024) 0 global_average_pooling2d[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, 1) 1025 dropout[0][0]
==================================================================================================
Total params: 2,782,649
Trainable params: 2,773,913
Non-trainable params: 8,736
__________________________________________________________________________________________________
None
짜잔! 이런 방법으로 도커환경을 이용하여 파이썬 파일을 실행해볼 수 있다.
끝
'Docker' 카테고리의 다른 글
| Pycharm과 Docker 연동 (2) | 2020.05.03 |
|---|
