mirror of
https://github.com/Tencent/Metis.git
synced 2025-12-26 04:02:48 +00:00
228 lines
9.6 KiB
Python
228 lines
9.6 KiB
Python
#!/usr/bin/env python
|
|
# -*- coding: UTF-8 -*-
|
|
"""
|
|
Tencent is pleased to support the open source community by making Metis available.
|
|
Copyright (C) 2018 THL A29 Limited, a Tencent company. All rights reserved.
|
|
Licensed under the BSD 3-Clause License (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
|
|
https://opensource.org/licenses/BSD-3-Clause
|
|
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.
|
|
"""
|
|
|
|
import numpy as np
|
|
from app.config.common import *
|
|
|
|
`
|
|
def time_series_moving_average(x):
|
|
"""
|
|
Returns the difference between the last element of x and the smoothed value after Moving Average Algorithm
|
|
The Moving Average Algorithm is M_{n} = (x_{n-w+1}+...+x_{n})/w, where w is a parameter
|
|
The parameter w is chosen in {1, 6, 11, 16, 21, 26, 31, 36, 41, 46} and the set of parameters can be changed.
|
|
|
|
WIKIPEDIA: https://en.wikipedia.org/wiki/Moving_average
|
|
|
|
:param x: the time series to calculate the feature of
|
|
:type x: pandas.Series
|
|
:return: the value of this feature
|
|
:return type: list with float
|
|
"""
|
|
temp_list = []
|
|
for w in range(1, min(50, DEFAULT_WINDOW), 5):
|
|
temp = np.mean(x[-w:])
|
|
temp_list.append(temp)
|
|
return list(np.array(temp_list) - x[-1])
|
|
|
|
|
|
def time_series_weighted_moving_average(x):
|
|
"""
|
|
Returns the difference between the last element of x and the smoothed value after Weighted Moving Average Algorithm
|
|
The Moving Average Algorithm is M_{n} = (1*x_{n-w+1}+...+(w-1)*x_{n-1}+w*x_{n})/w, where w is a parameter
|
|
The parameter w is chosen in {1, 6, 11, 16, 21, 26, 31, 36, 41, 46} and the set of parameters can be changed.
|
|
|
|
WIKIPEDIA: https://en.wikipedia.org/wiki/Moving_average
|
|
|
|
:param x: the time series to calculate the feature of
|
|
:type x: pandas.Series
|
|
:return: the value of this feature
|
|
:return type: list with float
|
|
"""
|
|
temp_list = []
|
|
for w in range(1, min(50, DEFAULT_WINDOW), 5):
|
|
w = min(len(x), w) # avoid the case len(value_list) < w
|
|
coefficient = np.array(range(1, w + 1))
|
|
temp_list.append((np.dot(coefficient, x[-w:])) / (w * (w + 1) / 2))
|
|
return list(np.array(temp_list) - x[-1])
|
|
|
|
|
|
def time_series_exponential_weighted_moving_average(x):
|
|
"""
|
|
Returns the difference between the last element of x and the smoothed value after Exponential Moving Average Algorithm
|
|
The Moving Average Algorithm is s[i] = alpha * x[i] + (1 - alpha) * s[i-1], where alpha is a parameter
|
|
The parameter w is chosen in {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9} and the set of parameters can be changed.
|
|
|
|
WIKIPEDIA: https://en.wikipedia.org/wiki/Exponential_smoothing
|
|
|
|
:param x: the time series to calculate the feature of
|
|
:type x: pandas.Series
|
|
:return: the value of this feature
|
|
:return type: list with float
|
|
"""
|
|
temp_list = []
|
|
for j in range(1, 10):
|
|
alpha = j / 10.0
|
|
s = [x[0]]
|
|
for i in range(1, len(x)):
|
|
temp = alpha * x[i] + (1 - alpha) * s[-1]
|
|
s.append(temp)
|
|
temp_list.append(s[-1] - x[-1])
|
|
return temp_list
|
|
|
|
|
|
def time_series_double_exponential_weighted_moving_average(x):
|
|
"""
|
|
Returns the difference between the last element of x and the smoothed value after Double Exponential Moving Average Algorithm
|
|
The Moving Average Algorithm is s[i] = alpha * x[i] + (1 - alpha) * (s[i-1] + b[i-1]), b[i] = gamma * (s[i] - s[i-1]) + (1 - gamma) * b[i-1]
|
|
where alpha and gamma are parameters.
|
|
The parameter alpha is chosen in {0.1, 0.3, 0.5, 0.7, 0.9} and the set of parameters can be changed.
|
|
The parameter gamma is chosen in {0.1, 0.3, 0.5, 0.7, 0.9} and the set of parameters can be changed.
|
|
|
|
WIKIPEDIA: https://en.wikipedia.org/wiki/Exponential_smoothing
|
|
|
|
:param x: the time series to calculate the feature of
|
|
:type x: pandas.Series
|
|
:return: the value of this feature
|
|
:return type: list with float
|
|
"""
|
|
temp_list = []
|
|
for j1 in range(1, 10, 2):
|
|
for j2 in range(1, 10, 2):
|
|
alpha = j1 / 10.0
|
|
gamma = j2 / 10.0
|
|
s = [x[0]]
|
|
b = [(x[3] - x[0]) / 3] # s is the smoothing part, b is the trend part
|
|
for i in range(1, len(x)):
|
|
temp1 = alpha * x[i] + (1 - alpha) * (s[-1] + b[-1])
|
|
s.append(temp1)
|
|
temp2 = gamma * (s[-1] - s[-2]) + (1 - gamma) * b[-1]
|
|
b.append(temp2)
|
|
temp_list.append(s[-1] - x[-1])
|
|
return temp_list
|
|
|
|
|
|
def time_series_periodic_features(data_c_left, data_c_right, data_b_left, data_b_right, data_a):
|
|
"""
|
|
Returns the difference between the last element of data_a and the last element of data_b_left,
|
|
the difference between the last element of data_a and the last element of data_c_left.
|
|
|
|
:param data_c_left: the time series of historical reference data
|
|
:type data_c_left: pandas.Series
|
|
:param data_c_right: the time series of historical reference data
|
|
:type data_c_right: pandas.Series
|
|
:param data_b_left: the time series of historical reference data
|
|
:type data_b_left: pandas.Series
|
|
:param data_b_right: the time series of historical reference data
|
|
:type data_b_right: pandas.Series
|
|
:param data_a: the time series to calculate the feature of
|
|
:type data_a: pandas.Series
|
|
:return: the value of this feature
|
|
:return type: list with float
|
|
"""
|
|
periodic_features = []
|
|
|
|
'''
|
|
Add the absolute value of difference between today and a week ago and its sgn as two features
|
|
Add the absolute value of difference between today and yesterday and its sgn as two features
|
|
'''
|
|
|
|
temp_value = data_c_left[-1] - data_a[-1]
|
|
periodic_features.append(abs(temp_value))
|
|
if temp_value < 0:
|
|
periodic_features.append(-1)
|
|
else:
|
|
periodic_features.append(1)
|
|
|
|
temp_value = data_b_left[-1] - data_a[-1]
|
|
periodic_features.append(abs(temp_value))
|
|
if temp_value < 0:
|
|
periodic_features.append(-1)
|
|
else:
|
|
periodic_features.append(1)
|
|
|
|
'''
|
|
If the last value of today is larger than the whole subsequence of a week ago,
|
|
then return the difference between the maximum of the whole subsequence of a week ago and the last value of today.
|
|
Others are similar.
|
|
'''
|
|
|
|
periodic_features.append(min(max(data_c_left) - data_a[-1], 0))
|
|
periodic_features.append(min(max(data_c_right) - data_a[-1], 0))
|
|
periodic_features.append(min(max(data_b_left) - data_a[-1], 0))
|
|
periodic_features.append(min(max(data_b_right) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_c_left) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_c_right) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_b_left) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_b_right) - data_a[-1], 0))
|
|
|
|
'''
|
|
If the last value of today is larger than the subsequence of a week ago,
|
|
then return the difference between the maximum of the whole subsequence of a week ago and the last value of today.
|
|
Others are similar.
|
|
'''
|
|
|
|
for w in range(1, DEFAULT_WINDOW, DEFAULT_WINDOW / 6):
|
|
periodic_features.append(min(max(data_c_left[-w:]) - data_a[-1], 0))
|
|
periodic_features.append(min(max(data_c_right[:w]) - data_a[-1], 0))
|
|
periodic_features.append(min(max(data_b_left[-w:]) - data_a[-1], 0))
|
|
periodic_features.append(min(max(data_b_right[:w]) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_c_left[-w:]) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_c_right[:w]) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_b_left[-w:]) - data_a[-1], 0))
|
|
periodic_features.append(max(min(data_b_right[:w]) - data_a[-1], 0))
|
|
|
|
'''
|
|
Add the difference of mean values between two subsequences
|
|
'''
|
|
|
|
for w in range(1, DEFAULT_WINDOW, DEFAULT_WINDOW / 9):
|
|
temp_value = np.mean(data_c_left[-w:]) - np.mean(data_a[-w:])
|
|
periodic_features.append(abs(temp_value))
|
|
if temp_value < 0:
|
|
periodic_features.append(-1)
|
|
else:
|
|
periodic_features.append(1)
|
|
|
|
temp_value = np.mean(data_c_right[:w]) - np.mean(data_a[-w:])
|
|
periodic_features.append(abs(temp_value))
|
|
if temp_value < 0:
|
|
periodic_features.append(-1)
|
|
else:
|
|
periodic_features.append(1)
|
|
|
|
temp_value = np.mean(data_b_left[-w:]) - np.mean(data_a[-w:])
|
|
periodic_features.append(abs(temp_value))
|
|
if temp_value < 0:
|
|
periodic_features.append(-1)
|
|
else:
|
|
periodic_features.append(1)
|
|
|
|
temp_value = np.mean(data_b_right[:w]) - np.mean(data_a[-w:])
|
|
periodic_features.append(abs(temp_value))
|
|
if temp_value < 0:
|
|
periodic_features.append(-1)
|
|
else:
|
|
periodic_features.append(1)
|
|
return periodic_features
|
|
|
|
# add yourself fitting features here...
|
|
|
|
|
|
def get_fitting_features(x_list):
|
|
fitting_features = []
|
|
fitting_features.extend(time_series_moving_average(x_list[4]))
|
|
fitting_features.extend(time_series_weighted_moving_average(x_list[4]))
|
|
fitting_features.extend(time_series_exponential_weighted_moving_average(x_list[4]))
|
|
fitting_features.extend(time_series_double_exponential_weighted_moving_average(x_list[4]))
|
|
fitting_features.extend(time_series_periodic_features(x_list[0], x_list[1], x_list[2], x_list[3], x_list[4]))
|
|
# append yourself fitting features here...
|
|
|
|
return fitting_features
|