728x90
반응형
Tensorflow
모델을 만드는 여러가지 표현 방법
1. Sequential
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Dense(2))
model.add(tf.keras.layers.Dense(2))
x = tf.constant([[1,2]])
y = tf.constant([[1,2],[2,3]])
model(x)
# <tf.Tensor: shape=(1, 2), dtype=float32, numpy=array([[0.3294789, 0.589738 ]], dtype=float32)>
model.built
# True
model.summary()
# Model: "sequential"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# dense (Dense) (1, 2) 6
# _________________________________________________________________
# dense_1 (Dense) (1, 2) 6
# =================================================================
# Total params: 12
# Trainable params: 12
# Non-trainable params: 0
# _________________________________________________________________model(y)
# <tf.Tensor: shape=(2, 2), dtype=float32, numpy=
# array([[0.3294789 , 0.589738 ],
# [0.2781678 , 0.97357416]], dtype=float32)>
model.summary()
# Model: "sequential"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# dense (Dense) (None, 2) 6
# _________________________________________________________________
# dense_1 (Dense) (None, 2) 6
# =================================================================
# Total params: 12
# Trainable params: 12
# Non-trainable params: 0
# _________________________________________________________________
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.InputLayer(input_shape=(2,))) # 두 개의 값을 갖는 1차원 데이터를 받겠다 (데이터의 개수는 상관없이 받는다)
model.add(tf.keras.layers.Dense(2))
model.add(tf.keras.layers.Dense(2))
model.summary()
# Model: "sequential_1"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# dense_2 (Dense) (None, 2) 6
# _________________________________________________________________
# dense_3 (Dense) (None, 2) 6
# =================================================================
# Total params: 12
# Trainable params: 12
# Non-trainable params: 0
# _________________________________________________________________model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Dense(2, input_shape=(2,))) # input layer를 추가하지 않고 줄여서 사용할 수 있다
model.add(tf.keras.layers.Dense(2))
model.summary()
# Model: "sequential_3"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# dense_5 (Dense) (None, 2) 6
# _________________________________________________________________
# dense_6 (Dense) (None, 2) 6
# =================================================================
# Total params: 12
# Trainable params: 12
# Non-trainable params: 0
# _________________________________________________________________model = tf.keras.models.Sequential([
tf.keras.layers.Dense(2, input_shape=[2]), # (2,)처럼 튜플 형태말고 리스트 형태로도 표현할 수 있다
tf.keras.layers.Dense(2)
])
model.summary()
# Model: "sequential_4"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# dense_7 (Dense) (None, 2) 6
# _________________________________________________________________
# dense_8 (Dense) (None, 2) 6
# =================================================================
# Total params: 12
# Trainable params: 12
# Non-trainable params: 0
# _________________________________________________________________Model
inputs = tf.keras.Input(shape=(2,)) # tensor
x = tf.keras.layers.Dense(2)(inputs)s
outputs = tf.keras.layers.Dense(2)(x)
model = tf.keras.models.Model(inputs, outputs) # input과 output을 같이 인자로 넣어줘야 한다 / 그리고 input으로 tensor를 받아야 한다
model.summary() # input layer가 명시된다
# Model: "model"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# input_4 (InputLayer) [(None, 2)] 0
# _________________________________________________________________
# dense_15 (Dense) (None, 2) 6
# _________________________________________________________________
# dense_16 (Dense) (None, 2) 6
# =================================================================
# Total params: 12
# Trainable params: 12
# Non-trainable params: 0
# _________________________________________________________________inputs = tf.keras.layers.InputLayer(input_shape=(2,)) # layer
x = tf.keras.layers.Dense(2)(inputs) # tensor를 인자로 받아야 한다
outputs = tf.keras.layers.Dense(2)(x)
# TypeError: Inputs to a layer should be tensors. Got: <keras.engine.input_layer.InputLayer object at 0x7f09f68a56d0># Sequential이 아닌 Model을 사용하는 것은 inputlayer를 표시하여서 조금더 복잡한 모델을 만들수있기 때문
inputs1 = tf.keras.Input(shape=(2,))
inputs2 = tf.keras.Input(shape=(2,))
x1 = tf.keras.layers.Dense(2)(inputs1)
x2 = tf.keras.layers.Dense(2)(inputs2)
outputs1 = tf.keras.layers.Dense(2)(x1)
outputs2 = tf.keras.layers.Dense(2)(x2)
model = tf.keras.models.Model([inputs1, inputs2], [outputs1,outputs2])
model.summary() # InputLayer나 connected to가 있을 경우 Model로 만들었다는 의미 이다
# Model: "model_1"
# __________________________________________________________________________________________________
# Layer (type) Output Shape Param # Connected to
# ==================================================================================================
# input_8 (InputLayer) [(None, 2)] 0
# __________________________________________________________________________________________________
# input_9 (InputLayer) [(None, 2)] 0
# __________________________________________________________________________________________________
# dense_19 (Dense) (None, 2) 6 input_8[0][0]
# __________________________________________________________________________________________________
# dense_20 (Dense) (None, 2) 6 input_9[0][0]
# __________________________________________________________________________________________________
# dense_21 (Dense) (None, 2) 6 dense_19[0][0]
# __________________________________________________________________________________________________
# dense_22 (Dense) (None, 2) 6 dense_20[0][0]
# ==================================================================================================
# Total params: 24
# Trainable params: 24
# Non-trainable params: 0
# __________________________________________________________________________________________________Sequential과 달리 input layer와 output layer를 두 개 이상 받을 수 있다
tf.keras.utils.plot_model(model)
tf.keras.utils.plot_model(model, show_shapes=True)
inputs1 = tf.keras.Input(shape=(2,))
inputs2 = tf.keras.Input(shape=(2,))
x1 = tf.keras.layers.Dense(2)(inputs1)
x2 = tf.keras.layers.Dense(2)(inputs2)
outputs1 = tf.keras.layers.Add()([x1,x2])
outputs2 = tf.keras.layers.Subtract()([x1,x2])
model = tf.keras.models.Model([inputs1, inputs2], [outputs1,outputs2])
tf.keras.utils.plot_model(model, show_shapes=True)
Model과 Sequential을 동시에 활용하는 방법
se = tf.keras.models.Sequential([
tf.keras.layers.Dense(2),
tf.keras.layers.Dense(2)
])
inputs_ = tf.keras.Input(shape=(3,))
outputs = se(inputs_)
model = tf.keras.models.Model(inputs_, outputs)
model
# <keras.engine.functional.Functional at 0x7f73508eef10>model.summary() # 모델 안에 모델을 넣지는 않지만 단순화 하기 위해서 모델 안에 sequential을 넣는 방법을 사용하는 경우는 많다
# Model: "model_2"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# input_6 (InputLayer) [(None, 3)] 0
# _________________________________________________________________
# sequential (Sequential) (None, 2) 14
# =================================================================
# Total params: 14
# Trainable params: 14
# Non-trainable params: 0
# _________________________________________________________________데이터 시각화로 Balanced data인지 구별하기
(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()
X_train.shape
# (60000, 28, 28)
X_train
# array([[[0, 0, 0, ..., 0, 0, 0],
# [0, 0, 0, ..., 0, 0, 0],
# [0, 0, 0, ..., 0, 0, 0],
# ...
# [0, 0, 0, ..., 0, 0, 0],
# [0, 0, 0, ..., 0, 0, 0],
# [0, 0, 0, ..., 0, 0, 0]]], dtype=uint8)np.unique(y_train, return_counts=True)
# (array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=uint8),
# array([5923, 6742, 5958, 6131, 5842, 5421, 5918, 6265, 5851, 5949]))
plt.hist(y_train) # blanced data
# (array([5923., 6742., 5958., 6131., 5842., 5421., 5918., 6265., 5851.,
# 5949.]),
# array([0. , 0.9, 1.8, 2.7, 3.6, 4.5, 5.4, 6.3, 7.2, 8.1, 9. ]),
# <a list of 10 Patch objects>)
MLP
# numerical stability => input이 있으면 output이 있어야 한다
# Normalization
X_train = X_train / 255 # coersion
X_test = X_test / 255preprocessing을 모델 안에서 하는지 모델 밖에서 하는지 결정해야 한다
Inside model
- portability, 편리하다
- GPU 연산이 가능하다
Outside model
- 데이터의 형태에 제약이 없어 범용성이 있다
- 기본적으로 CPU방식이지만 CPU + GPU방식을 쓸 수 있다
tensorflow에서는 3차원,즉 흑백 이미지는 (Batch(size), H, W)로 받아들인다
4차원(color 이미지) 일때는 (B, H, W, C) 또는 (B, C, H, W)(theano 기본)으로 받아들인다
Flatten
tf.keras.layers.Flatten()(X_train) # 함수 밖에서 사용하면 cpu로 연산한다
# <tf.Tensor: shape=(60000, 784), dtype=float32, numpy=
# array([[0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.],
# ...,
# [0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.]], dtype=float32)>
X_train_bw = X_train.reshape(60000, 28, 28, 1)
tf.keras.layers.Flatten()(X_train_bw)
# <tf.Tensor: shape=(60000, 784), dtype=float32, numpy=
# array([[0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.],
# ...,
# [0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.],
# [0., 0., 0., ..., 0., 0., 0.]], dtype=float32)>X_train_bw.flatten().shape
# (47040000,)
# 그냥 하면 depth 까지 다 flat해짐numpy에서 flatten은 모든 것을 1차원으로 만들어 준다
'__call__' in dir(tf.keras.layers.Flatten()) # instance => Function
# TrueMLP를 만드는 다양한 방법
input_ = tf.keras.Input(shape=(28,28))
x = tf.keras.layers.Flatten()(input_) # 1차가 되었기 때문에 fully conntected 모델을 사용할 수 있다
x = tf.keras.layers.Dense(128, activation='relu')(x) # Dense에는 하나의 데이터가 1차원인 경우에만 집어 넣을 수 있다
x
# <KerasTensor: shape=(None, 128) dtype=float32 (created by layer 'dense_6')>
input_ = tf.keras.Input(shape=(28,28))
x = tf.keras.layers.Dense(128, activation='relu')(input_) # 2차원 값을 Dense에 넣으면 원하는 output을 만들어 낼 수 없다
x
# <KerasTensor: shape=(None, 28, 128) dtype=float32 (created by layer 'dense_7')>Option1
input_ = tf.keras.Input(shape=(28,28))
x = tf.keras.layers.Flatten()(input_)
x = tf.keras.layers.Dense(2)(x)
x = tf.keras.layers.Activation('relu')(x) # 또는 x = tf.keras.layers.ReLU()(x)
# Activation layer를 따로 두는 이유는 Batch Normalization을 하기 위해서 이다Option2
input_ = tf.keras.Input(shape=(28,28))
x = tf.keras.layers.Flatten()(input_)
x= tf.keras.layers.Dense(128, activation='relu')(x)
input_ = tf.keras.Input(shape=(28,28))
x = tf.keras.layers.Flatten()(input_)
x = tf.keras.layers.Dense(128)(x)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.Dropout(0.2)(x)
output = tf.keras.layers.Dense(10, activation='softmax')(x)
# Batch Normalization과 상관 없기 때문에 관례상 마지막은 단축 표현을 쓴다
model = tf.keras.models.Model(input_, output)
model.summary() # param이 0인 경우 학습하지 않아도 되는 단순한 연산이다
# Model: "model_3"
# _________________________________________________________________
# Layer (type) Output Shape Param #
# =================================================================
# input_9 (InputLayer) [(None, 28, 28)] 0
# _________________________________________________________________
# flatten_1 (Flatten) (None, 784) 0
# _________________________________________________________________
# dense_8 (Dense) (None, 128) 100480
# _________________________________________________________________
# re_lu (ReLU) (None, 128) 0
# _________________________________________________________________
# dropout (Dropout) (None, 128) 0
# _________________________________________________________________
# dense_9 (Dense) (None, 10) 1290
# =================================================================
# Total params: 101,770
# Trainable params: 101,770
# Non-trainable params: 0
# _________________________________________________________________neural network parameter 구하는 법
입력값 (28x28) X dense layer nodes(128) + dense layer bias(128) = 100480
model.weights[0]
# tensorflow에서는 weight를 kernel라고 부른다 그리고 weight는 kernel + bias를 지칭한다
# kernel은 glorot방식으로 초기화 된다(학습이 빠르고 잘 수렴되는 초기화 방식)
# <tf.Variable 'dense_7/kernel:0' shape=(784, 128) dtype=float32, numpy=
# array([[-0.00481777, -0.05054575, -0.066313 , ..., 0.01311962,
# 0.04805059, 0.00298249],
# [ 0.07520033, -0.01051669, 0.00903524, ..., -0.07472851,
# 0.01202653, -0.00115251],
# [-0.01647546, -0.03879095, -0.03718614, ..., -0.05494369,
# -0.05540787, 0.05530912],
# ...,
# [-0.00182921, 0.07649232, -0.07937535, ..., 0.05838447,
# 0.05726648, -0.03632762],
# [-0.00923116, 0.01400795, -0.05262435, ..., -0.07948828,
# 0.01813819, -0.02076239],
# [-0.01461188, 0.07456786, 0.0081539 , ..., 0.01914987,
# 0.07928162, -0.03385995]], dtype=float32)>model.weights[1]
# <tf.Variable 'dense_7/bias:0' shape=(128,) dtype=float32, numpy=
# array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)>model.weights[2]
# <tf.Variable 'dense_8/kernel:0' shape=(128, 10) dtype=float32, numpy=
# array([[-0.01635328, 0.00763369, 0.13723023, ..., -0.18139066,
# 0.09355612, 0.14300655],
# [-0.06704196, -0.12000238, 0.20172156, ..., 0.0606968 ,
# -0.02551591, -0.10963563],
# [ 0.00977565, -0.11188473, 0.15327264, ..., 0.12097086,
# 0.00371699, 0.11089064],
# ...,
# [ 0.15975083, -0.12796631, -0.12143513, ..., 0.00048973,
# 0.08025642, 0.09352569],
# [-0.10482513, -0.00614406, 0.16832988, ..., 0.1809843 ,
# 0.16601638, -0.18317826],
# [ 0.10724227, 0.1758986 , 0.03089926, ..., 0.09623767,
# 0.0754603 , -0.14214656]], dtype=float32)>model.weights[3]
# <tf.Variable 'dense_8/bias:0' shape=(10,) dtype=float32, numpy=array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)>
model.trainable_weights == model.trainable_variables
# True
model.trainable_weights is model.trainable_variables
# False
model.trainable_weights
# [<tf.Variable 'dense_8/kernel:0' shape=(784, 128) dtype=float32, numpy=
# array([[-0.05531485, -0.07744511, -0.0746323 , ..., -0.06485563,
# -0.05913215, -0.03761471],
# [ 0.00124295, 0.07567873, 0.00529402, ..., 0.03713652,
# -0.03533567, 0.02690652],
# [ 0.06762456, -0.02606952, -0.03866001, ..., 0.0321365 ,
# -0.05224103, -0.03288466],
# ...,
# [ 0.05756753, 0.03487769, -0.04956114, ..., 0.07006838,
# -0.04104863, -0.08020123],
# [ 0.03739367, 0.05591244, 0.0753384 , ..., 0.0743489 ,
# 0.00566504, -0.05400074],
# [-0.00660357, -0.06026491, -0.07941656, ..., -0.05506966,
# -0.06525376, -0.05522396]], dtype=float32)>,
# <tf.Variable 'dense_8/bias:0' shape=(128,) dtype=float32, numpy=
# array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
# 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)>,
# <tf.Variable 'dense_9/kernel:0' shape=(128, 10) dtype=float32, numpy=
# array([[ 0.09544928, 0.08058016, -0.19936405, ..., 0.06292586,
# 0.05765642, -0.18936837],
# [-0.03744939, -0.09664733, 0.01312402, ..., 0.10993271,
# 0.14227311, 0.10375817],
# [-0.11351802, 0.05995773, 0.19267906, ..., 0.02832732,
# 0.1751137 , 0.08727919],
# ...,
# [ 0.18473722, 0.0153936 , 0.17687447, ..., -0.07043667,
# 0.07194577, -0.2060329 ],
# [ 0.04917909, -0.19079669, 0.03048648, ..., -0.04494686,
# 0.19417565, -0.18756153],
# [ 0.04652865, 0.02577874, -0.12224188, ..., -0.19071367,
# 0.04850109, 0.04275919]], dtype=float32)>,
# <tf.Variable 'dense_9/bias:0' shape=(10,) dtype=float32, numpy=array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)>]Learning
학습시키는 두 가지 방법
1. compile 방식 => computational graph 형태로 변환 시켜준다 / 내가 만든 모델이 tensorflow에서 지원해준다
2. 직접 구현하는 방식
반응형
'Computer_Science > Visual Intelligence' 카테고리의 다른 글
| 15일차 - CNN (0) | 2021.10.04 |
|---|---|
| 14일차 - 딥러닝 (Tensorflow) (3) (0) | 2021.10.04 |
| 12일차 - 딥러닝 (Tensorflow) (1) (0) | 2021.09.26 |
| 11일차 - 영상 데이터 기계학습 활용 (0) | 2021.09.26 |
| 9일차 - 영상 데이터 기계학습 활용 (0) | 2021.09.21 |