【GitHub】TensorFlow Tutorial for Time Series Prediction

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

TensorFlow Tutorial for Time Series Prediction

1. TensorFlow Tutorial for Time Series Prediction
   
2. 
   
3. This tutorial is designed to easily learn TensorFlow for time series prediction. Each tutorial subject includes both code and notebook with descriptions.
   
4. 
   
5. Tutorial Index
   
6. 
   
7. MNIST classification using Recurrent Neural Networks (RNN)
   
8. 
   
9. Classification for MNIST using RNN (notebook)
   
10. Time series prediction using Recurrent Neural Networks (RNN)
    
11. 
    
12. Prediction for sine wave function using Gaussian process (code / notebook)
    
13. Prediction for sine wave function using RNN (code / notebook)
    
14. Prediction for electricity price (code / notebook)
    
15. Slide materials
    
16. 
    
17. Slides on slideshare (TensorFlow-KR Meetup)
    
18. Slides on github (KSC 2016 Tutorial)

复制代码

本帖隐藏的内容

https://github.com/tgjeon/TensorFlow-Tutorials-for-Time-Series

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏2 回帖

关键词：Time Series Prediction Tutorial predict Tensor Series

1. # coding: utf-8
   
2. 
   
3. # ## Prediction sine wave function using Gaussian Process
   
4. #
   
5. # An example for Gaussian process algorithm to predict sine wave function.
   
6. # This example is from ["Gaussian Processes regression: basic introductory example"](http://scikit-learn.org/stable/auto_examples/gaussian_process/plot_gp_regression.html).
   
7. 
   
8. import numpy as np
   
9. from sklearn.gaussian_process import GaussianProcess
   
10. from matplotlib import pyplot as pl
    
11. %matplotlib inline
    
12. 
    
13. np.random.seed(1)
    
14. 
    
15. 
    
16. # The function to predict
    
17. def f(x):
    
18.     return x*np.sin(x)
    
19. 
    
20. 
    
21. # --------------------------
    
22. #  First the noiseless case
    
23. # --------------------------
    
24. 
    
25. # Obervations
    
26. X = np.atleast_2d([0., 1., 2., 3., 5., 6., 7., 8., 9.5]).T
    
27. y = f(X).ravel()
    
28. 
    
29. #X = np.atleast_2d(np.linspace(0, 100, 200)).T
    
30. 
    
31. # Mesh the input space for evaluations of the real function, the prediction and its MSE
    
32. x = np.atleast_2d(np.linspace(0, 10, 1000)).T
    
33. 
    
34. # Instanciate a Gaussian Process model
    
35. gp = GaussianProcess(corr='cubic', theta0=1e-2, thetaL=1e-4, thetaU=1e-1,
    
36.                      random_start=100)
    
37. 
    
38. # Fit to data using Maximum Likelihood Estimation of the parameters
    
39. gp.fit(X, y)
    
40. 
    
41. # Make the prediction on the meshed x-axis (ask for MSE as well)
    
42. y_pred, MSE = gp.predict(x, eval_MSE=True)
    
43. sigma = np.sqrt(MSE)
    
44. 
    
45. 
    
46. # Plot the function, the prediction and the 95% confidence interval based on the MSE
    
47. fig = pl.figure()
    
48. pl.plot(x, f(x), 'r:', label=u'$f(x) = x\,\sin(x)$')
    
49. pl.plot(X, y, 'r.', markersize=10, label=u'Observations')
    
50. pl.plot(x, y_pred, 'b-', label=u'Prediction')
    
51. pl.fill(np.concatenate([x, x[::-1]]),
    
52.         np.concatenate([y_pred - 1.9600 * sigma,
    
53.                        (y_pred + 1.9600 * sigma)[::-1]]),
    
54.         alpha=.5, fc='b', ec='None', label='95% confidence interval')
    
55. pl.xlabel('$x$')
    
56. pl.ylabel('$f(x)$')
    
57. pl.ylim(-10, 20)
    
58. pl.legend(loc='upper left')
    
59. 
    
60. 
    
61. # now the noisy case
    
62. X = np.linspace(0.1, 9.9, 20)
    
63. X = np.atleast_2d(X).T
    
64. 
    
65. # Observations and noise
    
66. y = f(X).ravel()
    
67. dy = 0.5 + 1.0 * np.random.random(y.shape)
    
68. noise = np.random.normal(0, dy)
    
69. y += noise
    
70. 
    
71. # Mesh the input space for evaluations of the real function, the prediction and
    
72. # its MSE
    
73. x = np.atleast_2d(np.linspace(0, 10, 1000)).T
    
74. 
    
75. # Instanciate a Gaussian Process model
    
76. gp = GaussianProcess(corr='squared_exponential', theta0=1e-1,
    
77.                      thetaL=1e-3, thetaU=1,
    
78.                      nugget=(dy / y) ** 2,
    
79.                      random_start=100)
    
80. 
    
81. # Fit to data using Maximum Likelihood Estimation of the parameters
    
82. gp.fit(X, y)
    
83. 
    
84. # Make the prediction on the meshed x-axis (ask for MSE as well)
    
85. y_pred, MSE = gp.predict(x, eval_MSE=True)
    
86. sigma = np.sqrt(MSE)
    
87. 
    
88. # Plot the function, the prediction and the 95% confidence interval based on
    
89. # the MSE
    
90. fig = pl.figure()
    
91. pl.plot(x, f(x), 'r:', label=u'$f(x) = x\,\sin(x)$')
    
92. pl.errorbar(X.ravel(), y, dy, fmt='r.', markersize=10, label=u'Observations')
    
93. pl.plot(x, y_pred, 'b-', label=u'Prediction')
    
94. pl.fill(np.concatenate([x, x[::-1]]),
    
95.         np.concatenate([y_pred - 1.9600 * sigma,
    
96.                        (y_pred + 1.9600 * sigma)[::-1]]),
    
97.         alpha=.5, fc='b', ec='None', label='95% confidence interval')
    
98. pl.xlabel('$x$')
    
99. pl.ylabel('$f(x)$')
    
100. pl.ylim(-10, 20)
     
101. pl.legend(loc='upper left')
     
102. 
     
103. pl.show()

复制代码

1. # coding: utf-8
   
2. 
   
3. # In[13]:
   
4. 
   
5. # get_ipython().magic('matplotlib inline')
   
6. import numpy as np
   
7. from matplotlib import pyplot as plt
   
8. 
   
9. from tensorflow.contrib import learn
   
10. from sklearn.metrics import mean_squared_error, mean_absolute_error
    
11. from lstm_predictor import generate_data, lstm_model
    
12. 
    
13. 
    
14. # ## Libraries
    
15. #
    
16. # - numpy: package for scientific computing
    
17. # - pandas: data structures and data analysis tools
    
18. # - tensorflow: open source software library for machine intelligence
    
19. # - matplotlib: 2D plotting library
    
20. #
    
21. #
    
22. # - **learn**: Simplified interface for TensorFlow (mimicking Scikit Learn) for Deep Learning
    
23. # - mse: "mean squared error" as evaluation metric
    
24. # - **lstm_predictor**: our lstm class
    
25. #
    
26. 
    
27. # In[14]:
    
28. 
    
29. LOG_DIR = './ops_logs'
    
30. TIMESTEPS = 5
    
31. RNN_LAYERS = [{'steps': TIMESTEPS}]
    
32. DENSE_LAYERS = [10, 10]
    
33. TRAINING_STEPS = 100000
    
34. BATCH_SIZE = 100
    
35. PRINT_STEPS = TRAINING_STEPS / 100
    
36. 
    
37. 
    
38. # ## Parameter definitions
    
39. #
    
40. # - LOG_DIR: log file
    
41. # - TIMESTEPS: RNN time steps
    
42. # - RNN_LAYERS: RNN layer 정보
    
43. # - DENSE_LAYERS: DNN 크기 [10, 10]: Two dense layer with 10 hidden units
    
44. # - TRAINING_STEPS: 학습 스텝
    
45. # - BATCH_SIZE: 배치 학습 크기
    
46. # - PRINT_STEPS: 학습 과정 중간 출력 단계 (전체의 1% 해당하는 구간마다 출력)
    
47. 
    
48. # In[15]:
    
49. 
    
50. regressor = learn.TensorFlowEstimator(model_fn=lstm_model(TIMESTEPS, RNN_LAYERS, DENSE_LAYERS),
    
51.                                       n_classes=0,
    
52.                                       verbose=1,  
    
53.                                       steps=TRAINING_STEPS,
    
54.                                       optimizer='Adagrad',
    
55.                                       learning_rate=0.03,
    
56.                                       batch_size=BATCH_SIZE)
    
57. 
    
58. 
    
59. # ## Create a regressor with TF Learn
    
60. #
    
61. # : 예측을 위한 모델 생성. TF learn 라이브러리에 제공되는 TensorFlowEstimator를 사용.
    
62. #
    
63. # **Parameters**:
    
64. #
    
65. # - model_fn: 학습 및 예측에 사용할 모델
    
66. # - n_classes: label에 해당하는 클래스 수 (0: prediction, 1이상: classification) 확인필요
    
67. # - verbose: 과정 출력
    
68. # - steps: 학습 스텝
    
69. # - optimizer: 최적화 기법 ("SGD", "Adam", "Adagrad")
    
70. # - learning_rate: learning rate
    
71. # - batch_size: batch size
    
72. #
    
73. #
    
74. 
    
75. # In[ ]:
    
76. 
    
77. X, y = generate_data(np.sin, np.linspace(0, 100, 10000), TIMESTEPS, seperate=False)
    
78. # create a lstm instance and validation monitor
    
79. 
    
80. validation_monitor = learn.monitors.ValidationMonitor(X['val'], y['val'],
    
81.                                                       every_n_steps=PRINT_STEPS,
    
82.                                                       early_stopping_rounds=1000)
    
83. 
    
84. 
    
85. # ## Generate a dataset
    
86. #
    
87. # 1. generate_data: 학습에 사용될 데이터를 특정 함수를 이용하여 만듭니다.
    
88. #  - fct: 데이터를 생성할 함수
    
89. #  - x: 함수값을 관측할 위치
    
90. #  - time_steps: 관측(observation)
    
91. #  - seperate: check multimodal
    
92. # 1. ValidationMonitor: training 이후, validation 과정을 모니터링
    
93. #  - x
    
94. #  - y
    
95. #  - every_n_steps: 중간 출력
    
96. #  - early_stopping_rounds
    
97. 
    
98. # In[16]:
    
99. 
    
100. regressor.fit(X['train'], y['train'], monitors=[validation_monitor], logdir=LOG_DIR)
     
101. 
     
102. 
     
103. # ## Train and validation
     
104. #
     
105. # - fit: training data를 이용해 학습, 모니터링과 로그를 통해 기록
     
106. #
     
107. #
     
108. 
     
109. # In[17]:
     
110. 
     
111. predicted = regressor.predict(X['test'])
     
112. mse = mean_squared_error(y['test'], predicted)
     
113. print ("Error: %f" % mse)
     
114. 
     
115. 
     
116. # ## Evaluate using test set
     
117. #
     
118. # Evaluate our hypothesis using test set. The mean squared error (MSE) is used for the evaluation metric.
     
119. #
     
120. #
     
121. 
     
122. # In[18]:
     
123. 
     
124. plot_predicted, = plt.plot(predicted, label='predicted')
     
125. plot_test, = plt.plot(y['test'], label='test')
     
126. plt.legend(handles=[plot_predicted, plot_test])
     
127. 
     
128. 
     
129. 
     
130. # ## Plotting
     
131. #
     
132. # Then, plot both predicted values and original values from test set.

复制代码

1. import numpy as np
   
2. import pandas as pd
   
3. from matplotlib import pyplot as plt
   
4. 
   
5. from tensorflow.contrib import learn
   
6. from sklearn.metrics import mean_squared_error, mean_absolute_error
   
7. from lstm_predictor import generate_data, load_csvdata, lstm_model
   
8. 
   
9. 
   
10. LOG_DIR = './ops_logs'
    
11. TIMESTEPS = 10
    
12. RNN_LAYERS = [{'steps': TIMESTEPS}]
    
13. DENSE_LAYERS = [10, 10]
    
14. TRAINING_STEPS = 100000
    
15. BATCH_SIZE = 100
    
16. PRINT_STEPS = TRAINING_STEPS / 100
    
17. 
    
18. dateparse = lambda dates: pd.datetime.strptime(dates, '%d/%m/%Y %H:%M')
    
19. rawdata = pd.read_csv("./input/ElectricityPrice/RealMarketPriceDataPT.csv",
    
20.                    parse_dates={'timeline': ['date', '(UTC)']},
    
21.                    index_col='timeline', date_parser=dateparse)
    
22. 
    
23. 
    
24. X, y = load_csvdata(rawdata, TIMESTEPS, seperate=False)
    
25. 
    
26. 
    
27. regressor = learn.TensorFlowEstimator(model_fn=lstm_model(TIMESTEPS, RNN_LAYERS, DENSE_LAYERS),
    
28.                                       n_classes=0,
    
29.                                       verbose=1,  
    
30.                                       steps=TRAINING_STEPS,
    
31.                                       optimizer='Adagrad',
    
32.                                       learning_rate=0.03,
    
33.                                       batch_size=BATCH_SIZE)
    
34. 
    
35. 
    
36. 
    
37. 
    
38. validation_monitor = learn.monitors.ValidationMonitor(X['val'], y['val'],
    
39.                                                       every_n_steps=PRINT_STEPS,
    
40.                                                       early_stopping_rounds=1000)
    
41. 
    
42. regressor.fit(X['train'], y['train'], monitors=[validation_monitor], logdir=LOG_DIR)
    
43. 
    
44. 
    
45. predicted = regressor.predict(X['test'])
    
46. mse = mean_absolute_error(y['test'], predicted)
    
47. print ("Error: %f" % mse)
    
48. 
    
49. plot_predicted, = plt.plot(predicted, label='predicted')
    
50. plot_test, = plt.plot(y['test'], label='test')
    
51. plt.legend(handles=[plot_predicted, plot_test])

复制代码

谢谢楼主分享！

thanks for sharing

谢谢楼主分享

good sharing!

感谢楼主分享