Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
tensorflow
GitHub Repository: tensorflow/docs-l10n
Path: blob/master/site/ko/guide/variable.ipynb
25115 views
Kernel: Python 3
#@title Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # https://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License.

변수 소개

TensorFlow.org에서 보기 Google Colab에서 실행 GitHub에서 소스 보기 노트북 다운로드

TensorFlow 변수는 프로그램이 조작하는 공유 영구 상태를 나타내는 권장 방법입니다. 이 가이드는 TensorFlow에서 tf.Variable 인스턴스를 작성, 업데이트 및 관리하는 방법을 설명합니다.

변수는 tf.Variable 클래스를 통해 생성 및 추적됩니다. tf.Variable은 ops를 실행하여 값을 변경할 수 있는 텐서를 나타냅니다. 특정 ops를 사용하면 이 텐서의 값을 읽고 수정할 수 있습니다. tf.keras와 같은 상위 수준의 라이브러리는 tf.Variable을 사용하여 모델 매개변수를 저장합니다.

설정

변수 배치에 대해 설명하는 노트북입니다. 변수가 어떤 기기에 배치되었는지 보려면 이 줄의 주석을 해제하세요.

import tensorflow as tf # Uncomment to see where your variables get placed (see below) # tf.debugging.set_log_device_placement(True)

변수 만들기

변수를 작성하려면 초기값을 제공합니다. tf.Variable은 초기화 값과 같은 dtype을 갖습니다.

my_tensor = tf.constant([[1.0, 2.0], [3.0, 4.0]]) my_variable = tf.Variable(my_tensor) # Variables can be all kinds of types, just like tensors bool_variable = tf.Variable([False, False, False, True]) complex_variable = tf.Variable([5 + 4j, 6 + 1j])

변수는 텐서처럼 보이고 작동하며, 실제로 tf.Tensor에서 지원되는 데이터 구조입니다. 텐서와 마찬가지로, dtype과 형상을 가지며 NumPy로 내보낼 수 있습니다.

print("Shape: ", my_variable.shape) print("DType: ", my_variable.dtype) print("As NumPy: ", my_variable.numpy())

변수를 재구성할 수는 없지만, 대부분의 텐서 연산은 예상대로 변수에 대해 작동합니다.

print("A variable:", my_variable) print("\nViewed as a tensor:", tf.convert_to_tensor(my_variable)) print("\nIndex of highest value:", tf.math.argmax(my_variable)) # This creates a new tensor; it does not reshape the variable. print("\nCopying and reshaping: ", tf.reshape(my_variable, [1,4]))

위에서 언급했듯이, 변수는 텐서에 의해 지원됩니다. tf.Variable.assign을 사용하여 텐서를 재할당할 수 있습니다. assign을 호출해도 (일반적으로) 새로운 텐서를 할당하지 않고, 대신 기존 텐서의 메모리가 재사용됩니다.

a = tf.Variable([2.0, 3.0]) # This will keep the same dtype, float32 a.assign([1, 2]) # Not allowed as it resizes the variable: try: a.assign([1.0, 2.0, 3.0]) except Exception as e: print(f"{type(e).__name__}: {e}")

연산에서 텐서와 같은 변수를 사용하는 경우, 일반적으로 지원 텐서에서 작동합니다.

기존 변수에서 새 변수를 만들면 지원 텐서가 복제됩니다. 두 변수는 같은메모리를 공유하지 않습니다.

a = tf.Variable([2.0, 3.0]) # Create b based on the value of a b = tf.Variable(a) a.assign([5, 6]) # a and b are different print(a.numpy()) print(b.numpy()) # There are other versions of assign print(a.assign_add([2,3]).numpy()) # [7. 9.] print(a.assign_sub([7,9]).numpy()) # [0. 0.]

수명 주기, 이름 지정 및 감시

파이썬 기반 TensorFlow에서 tf.Variable 인스턴스는 다른 Python 객체와 같은 수명 주기를 갖습니다. 변수에 대한 참조가 없으면 자동으로 할당이 해제됩니다.

변수를 추적하고 디버그하는 데 도움이 되는 변수의 이름을 지정할 수도 있습니다. 두 변수에 같은 이름을 지정할 수 있습니다.

# Create a and b; they will have the same name but will be backed by # different tensors. a = tf.Variable(my_tensor, name="Mark") # A new variable with the same name, but different value # Note that the scalar add is broadcast b = tf.Variable(my_tensor + 1, name="Mark") # These are elementwise-unequal, despite having the same name print(a == b)

모델을 저장하고 로드할 때 변수 이름이 유지됩니다. 기본적으로 모델의 변수는 고유한 변수 이름이 자동으로 지정되므로 원치 않는 한 직접 할당할 필요가 없습니다.

변수는 구별을 위해 중요하지만, 일부 변수는 구별할 필요가 없습니다. 생성 시 trainable을 false로 설정하여 변수의 그래디언트를 끌 수 있습니다. 그래디언트가 필요하지 않은 변수의 예는 훈련 단계 카운터입니다.

step_counter = tf.Variable(1, trainable=False)

변수 및 텐서 배치하기

더 나은 성능을 위해 TensorFlow는 dtype과 호환되는 가장 빠른 기기에 텐서 및 변수를 배치하려고 시도합니다. 이는 대부분의 변수가 GPU(사용 가능한 경우)에 배치됨을 의미합니다.

그러나 우리는 이것을 재정의할 수 있습니다. 이 코드에서, GPU가 사용가능함에서도 우리는 부동 텐서와 변수를 CPU에 배치할 수 있습니다. 기기배치에 로그인을 함으로써 ( Setup 참조), 변수가 어디에 위치해있는지 확인할 수 있습니다.

참고: 수동 배치도 가능하지만, 분배 전략을 사용하여 더 편리하고 확장 가능하게 계산을 최적화 할 수 있습니다.

GPU가 있거나 없는 서로 다른 백엔드에서 이 노트북을 실행하면 서로 다른 로깅이 표시됩니다. 세션 시작 시 기기 배치 로깅을 켜야 합니다.

with tf.device('CPU:0'): # Create some tensors a = tf.Variable([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) b = tf.constant([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]) c = tf.matmul(a, b) print(c)

한 기기에서 변수 또는 텐서의 위치를 설정하고 다른 기기에서 계산을 수행할 수 있습니다. 이 경우, 기기 간에 데이터를 복사해야 하므로 지연이 발생합니다.

GPU 작업자가 여러 개이지만 변수의 사본이 하나만 필요한 경우에 수행할 수 있습니다.

with tf.device('CPU:0'): a = tf.Variable([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) b = tf.Variable([[1.0, 2.0, 3.0]]) with tf.device('GPU:0'): # Element-wise multiply k = a * b print(k)

주의: tf.config.set_soft_device_placement 이 기본적으로 항상 켜져있기 때문에, GPU가 없는 기기에서 이 코드를 실행하더라도 코드는 계속 실행됩니다. 곱셈 단계는 CPU에서 실행됩니다.

분산 훈련에 대한 자세한 내용은 가이드를 참조하세요.

다음 단계

변수들이 일반적으로 사용되는 방식을 이해하려면, automatic differentiation에서 우리의 가이드를 참조하세요.