Notice

Recent Posts

Recent Comments

Link

« 2025/05 »
일	월	화	수	목	금	토
				1	2	3
4	5	6	7	8	9	10
11	12	13	14	15	16	17
18	19	20	21	22	23	24
25	26	27	28	29	30	31

Tags more

Archives

Today

Total

관리 메뉴

Big Bro's Studying Archive

텐서플로2 기초 02) 다변량 선형 회귀 간단하게 구현해보기 본문

Big Data展

텐서플로2 기초 02) 다변량 선형 회귀 간단하게 구현해보기

빅브로오 2020. 9. 21. 17:37

TensorFlow2에서 다변량 선형 회귀 간단하게 구현해보기

# Multi variable linear regression
import tensorflow as tf
import numpy as np

인스턴스가 5이고 변수가 세개인 데이터 생성
이전 세 쪽지 시험 결과로 200점 만점의 기말고사 결과 예측

x_data = [[73., 80., 75.],
          [93., 88., 93.],
          [89., 91., 90.],
          [96., 98., 100.],
          [73., 66., 70.]]
y_data = [[152.],
          [185.],
          [180.],
          [196.],
          [142.]]

keras neural net 모델 설계하기

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Dense(units = 1, input_dim = 3, activation = 'linear'))

변수가 3개이기 때문에 input을 세개를 받는 모델을 설계해준다.
활성함수는 디폴트가 선형이기 때문에 선형회귀의 경우 최종 노드에서 별도로 설정해줄 필요는 사실 없다.

model compile 하기

여기서는 loss 함수와 옵티마이저를 설정해준다.

model.compile(loss = 'mse', optimizer = tf.keras.optimizers.SGD(lr = 1e-5))
model.summary() # 모델 확인

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 1)                 4         
=================================================================
Total params: 4
Trainable params: 4
Non-trainable params: 0
_________________________________________________________________

하나의 결과만을 출력하는 모델이 설계되었다.
학습시켜보자.

hist = model.fit(x_data, y_data, epochs = 100)

Train on 5 samples
Epoch 1/100
5/5 [==============================] - 0s 84ms/sample - loss: 93357.6016
Epoch 2/100
5/5 [==============================] - 0s 2ms/sample - loss: 29263.9375
Epoch 3/100
5/5 [==============================] - 0s 2ms/sample - loss: 9173.9766
Epoch 4/100
5/5 [==============================] - 0s 2ms/sample - loss: 2876.8394
Epoch 5/100
5/5 [==============================] - 0s 2ms/sample - loss: 903.0208
Epoch 6/100
5/5 [==============================] - 0s 2ms/sample - loss: 284.3332
Epoch 7/100
5/5 [==============================] - 0s 2ms/sample - loss: 90.4071
Epoch 8/100
5/5 [==============================] - 0s 2ms/sample - loss: 29.6211
Epoch 9/100
5/5 [==============================] - 0s 2ms/sample - loss: 10.5672
Epoch 10/100
5/5 [==============================] - 0s 2ms/sample - loss: 4.5944
Epoch 11/100
5/5 [==============================] - 0s 2ms/sample - loss: 2.7217
Epoch 12/100
5/5 [==============================] - 0s 2ms/sample - loss: 2.1343
Epoch 13/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.9496
Epoch 14/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8912
Epoch 15/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8724
Epoch 16/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8660
Epoch 17/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8635
Epoch 18/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8621
Epoch 19/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8612
Epoch 20/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8604
Epoch 21/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8597
Epoch 22/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8589
Epoch 23/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8582
Epoch 24/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8574
Epoch 25/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8567
Epoch 26/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8559
Epoch 27/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8552
Epoch 28/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8545
Epoch 29/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8537
Epoch 30/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8530
Epoch 31/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8522
Epoch 32/100
5/5 [==============================] - 0s 4ms/sample - loss: 1.8515
Epoch 33/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8507
Epoch 34/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8500
Epoch 35/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8493
Epoch 36/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8485
Epoch 37/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8478
Epoch 38/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8471
Epoch 39/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8463
Epoch 40/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8456
Epoch 41/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8448
Epoch 42/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8441
Epoch 43/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8434
Epoch 44/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8426
Epoch 45/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8419
Epoch 46/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8412
Epoch 47/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8404
Epoch 48/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8397
Epoch 49/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8390
Epoch 50/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8382
Epoch 51/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8375
Epoch 52/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8368
Epoch 53/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8360
Epoch 54/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8353
Epoch 55/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8346
Epoch 56/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8338
Epoch 57/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8331
Epoch 58/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8324
Epoch 59/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8316
Epoch 60/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8309
Epoch 61/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8302
Epoch 62/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8295
Epoch 63/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8287
Epoch 64/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8280
Epoch 65/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8273
Epoch 66/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8266
Epoch 67/100
5/5 [==============================] - 0s 1ms/sample - loss: 1.8258
Epoch 68/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8251
Epoch 69/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8244
Epoch 70/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8237
Epoch 71/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8229
Epoch 72/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8222
Epoch 73/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8215
Epoch 74/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8208
Epoch 75/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8200
Epoch 76/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8193
Epoch 77/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8186
Epoch 78/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8179
Epoch 79/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8172
Epoch 80/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8164
Epoch 81/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8157
Epoch 82/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8150
Epoch 83/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8143
Epoch 84/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8135
Epoch 85/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8128
Epoch 86/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8121
Epoch 87/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8114
Epoch 88/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8107
Epoch 89/100
5/5 [==============================] - 0s 1ms/sample - loss: 1.8100
Epoch 90/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8092
Epoch 91/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8085
Epoch 92/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8078
Epoch 93/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8071
Epoch 94/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8064
Epoch 95/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8057
Epoch 96/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8049
Epoch 97/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8042
Epoch 98/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8035
Epoch 99/100
5/5 [==============================] - 0s 3ms/sample - loss: 1.8028
Epoch 100/100
5/5 [==============================] - 0s 2ms/sample - loss: 1.8021

y_pred = model.predict(np.array([[72.,93.,90.]]))
print(y_pred)

[[172.32043]]

이 모델은 약 172점을 예측했다.

hist.params

{'batch_size': 32,
 'epochs': 100,
 'steps': 1,
 'samples': 5,
 'verbose': 1,
 'do_validation': False,
 'metrics': ['loss']}

선형 회귀의 기초를 재현해보는 내용이기 때문에 별도의 파라미터값 설정 필요가 없지만,
Neural Net을 활용할 경우 파라미터 값에 유의해야 한다.

위 모델은 별도의 val data 없이 100 에폭을 학습시킨 모델이다.

'Big Data展' 카테고리의 다른 글

텐서플로2 기초 03) Logistic Classification 간단하게 구현하기 (0)	2020.09.22
tensorflow2에서 손실함수 동작 원리 실습해보기 (0)	2020.09.21
SQL 기초 : 다중 컬럼 서브쿼리 : 중첩서브쿼리와 상호연관 서브쿼리, 그리고 인라인 뷰 (Nested subquery, Correlated subquery, Inline view) (0)	2020.05.12
SQL 기초 : 다중행 서브쿼리의 사용법 (0)	2020.05.11
Python 기초 7 : 전역변수 선언, 사용자 설정 함수 인자, open함수 등 (0)	2020.05.08

'Big Data展' Related Articles

Comments

Big Bro's Studying Archive

텐서플로2 기초 02) 다변량 선형 회귀 간단하게 구현해보기 본문

텐서플로2 기초 02) 다변량 선형 회귀 간단하게 구현해보기

TensorFlow2에서 다변량 선형 회귀 간단하게 구현해보기

'Big Data展' 카테고리의 다른 글

티스토리툴바