LightGBM_timeseries_1d.py

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import lightgbm as lgb
from sklearn.model_selection import *


!wget "https://www.analyticsvidhya.com/wp-content/uploads/2016/02/AirPassengers.csv"


df = pd.read_csv('AirPassengers.csv', index_col=0)
f_ori = df.values
f = f_ori


length_of_sequences = len(df)
width = 12

data = []
target = []

for i in range(0, length_of_sequences - width + 1):
    data.append(f[i: i + width])
    target.append(f[i + width - 1])
    
    
X = np.stack(data, axis=0).reshape(-1, width)
y = np.array(target).reshape(len(data), )
X_train, X_validation, y_train, y_validation = train_test_split(X, y, test_size=0.1)


lgb_train = lgb.Dataset(X_train, y_train)
lgb_eval = lgb.Dataset(X_validation, y_validation, reference=lgb_train)

# LightGBM parameters
params = {
        'task' : 'train',
        'boosting':'gbdt',
        'objective' : 'regression',  # 7.0713
        'metric' : {'mae'},
        'num_leaves':78,
        'max_drop':20,  # 5
        'drop_rate':0.05,  # 0.01
        'learning_rate':0.01,
        'feature_fraction':0.4,  # 0.7
        'bagging_fraction':0.4,  # 0.85
        'bagging_freq':30,
        'lambda_l1':1,  # 1
        'lambda_l2':0.01,  # 0.01
        'seed':0,
        'verbose':0,
        'device': 'cpu'
}

evaluation_results = {}
gbm = lgb.train(params,
                lgb_train,
                num_boost_round=100000,
                valid_sets=[lgb_train, lgb_eval],
                valid_names=['Train', 'Test'],
                evals_result=evaluation_results,
                early_stopping_rounds=3000,
                verbose_eval=100)

y_pred = gbm.predict(X_validation, num_iteration=gbm.best_iteration)


# widthブロックの1個目を取得
Z = X[:1]  # X[0]としないのはshapeを(1, width)とするため
original = [f[i] for i in range(width)]
predicted = [None for i in range(width)]

for i in range(length_of_sequences - width + 1):
    z_ = Z[-1:]  # shape=(1, width, 1)
    y_ = gbm.predict(z_, num_iteration=gbm.best_iteration)
    sequence_ = np.concatenate((z_.reshape(width)[1:], y_), axis=0).reshape(1, width)
    Z = np.append(Z, sequence_, axis=0)
    predicted.append(y_.reshape(-1))
    
    
plt.plot(original, linestyle='dashed', color='black')
plt.plot(predicted, color='black')


# trainig dataに対する予測
pred_train = np.array([None for i in range(width)])
pred_train = np.append(pred_train, gbm.predict(X), axis=0)

plt.plot(f, linestyle='dashed', color='green')
plt.plot(original, linestyle='dashed', color='black')
plt.plot(pred_train, color='black')