[ML] Multi-step CNN Models

GITHUB

https://github.com/yiwonjae/Project_Lotto/blob/master/Book_001/p111.py

 

0. 목표

output이 여러단계를 원할 때

 

1. DATA

raw_seq = [10, 20, 30, 40, 50, 60, 70, 80, 90]

 

2. DATA 정제

X

from numpy import ndarray
import numpy as np

def split_sequence(sequence:list, n_steps_in:int, n_steps_out:int)->(ndarray, ndarray):

    x = []
    y = []

    for i in range(len(sequence)):

        if(i+n_steps_in+n_steps_out > len(sequence)):
            break

        x.append(sequence[i:i+n_steps_in])
        y.append(sequence[i+n_steps_in : i+n_steps_in + n_steps_out])


    return (np.asarray(x), np.asarray(y))

n_steps_in  = 3
n_steps_out = 2

(x, y) = split_sequence(raw_seq, n_steps_in, n_steps_out)


print(x.shape)
print(y.shape)

n_feature = 1

x = x.reshape((x.shape[0], x.shape[1], n_feature))

 

3. 학습

from keras import Sequential
from keras.layers import Dense, Conv1D, MaxPool1D, Flatten

model = Sequential()
model.add(Conv1D(64, 2, activation='relu', input_shape=(n_steps_in, n_feature)))
model.add(MaxPool1D())
model.add(Flatten())
model.add(Dense(50, activation='relu'))
model.add(Dense(n_steps_out))
model.compile(optimizer='adam', loss='mse')

model.fit(x, y, epochs=2000, verbose=1)

 

4. 표시

x_input = np.asarray([70, 80, 90])
x_input = x_input.reshape((1, n_steps_in, n_feature))

yhat = model.predict(x_input, verbose=1) # expect : [100, 110]
print(yhat) #[[102.19693 113.60678]]