TimeSeriesGenerator.ipynb

TimeSeriesGenerator.ipynb

# https://machinelearningmastery.com/how-to-use-the-timeseriesgenerator-for-time-series-forecasting-in-keras/
# univariate one step problem with lstm
from numpy import array
from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.layers import BatchNormalization, Dropout

from keras.layers import LSTM
from keras.preprocessing.sequence import TimeseriesGenerator

# define dataset
#series = array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
import pandas as pd 
import numpy as np
from google.colab import drive 
drive.mount('/content/gdrive')
Streamflow=pd.read_csv('gdrive/My Drive/NallihanStandardize.csv', skiprows=1, infer_datetime_format=True, names = ["Orj","S1","S2","S3","S5-1-1","S50-1"]) 
series1= Streamflow['Orj']
series=np.array(series1)
print (series[0:5])

n_data=len(series)
n_train=int(len(series) * 0.7)#4857 #699
n_test=int(len(series) * 0.2)#1387 #201 
n_valid=n_data-(n_train+n_test)#695 #100

# reshape to [10, 1]
n_features = 1 #univariate
series = series.reshape((len(series), n_features))
pred=series[0:1]
print('measured:',len(series),'-->',series.shape)

# define generator
n_input = 1
generator = TimeseriesGenerator(series, series, length=n_input, batch_size=1)
# define model
model = Sequential()
model.add(LSTM(160, activation='relu', input_shape=(n_input, n_features)))
model.add(Dense(1))
#model.add(LSTM(50, activation='relu'))
model.add(Dropout(0.10))
model.compile(optimizer='adam', loss='mse')
# fit model
for nfit in range(125):
  model.fit_generator(generator, steps_per_epoch=1, epochs=1, verbose=0, shuffle=False)
  model.reset_states()
# make a one step prediction out of sample
for n_phase in range(0,len(series)-n_input-n_valid):
  x_input = series[n_phase+0: n_phase+n_input].reshape((1, n_input, n_features)) # Bu aralık ile n_input = 3 eşit olmalı
  yhat = model.predict(x_input, verbose=0)
  model.reset_states()
  pred=np.append (pred, yhat)
pred=pred[1:len(pred)-1]
print('predicted:',len(pred),'--',pred.shape)

from matplotlib import pyplot
series=np.append (series, -5)
series1=np.append (series1, 5)

pyplot.figure(figsize=(12, 6))
pyplot.plot(series, label='Measured')
pyplot.plot(series1, label='Measured1')

pyplot.plot(pred, label='predicted')
pyplot.xlabel('Day')
pyplot.ylabel('Streamflow')
pyplot.title ('All predicteds')
pyplot.legend()
pyplot.show()

pyplot.figure(figsize=(12, 6))
pyplot.plot(series[400:600], label='Measured')
pyplot.plot(pred[400:600], label='predicted')
pyplot.xlabel('Day')
pyplot.ylabel('Streamflow')
pyplot.title ('Predicteds (400-600)')
pyplot.legend()
pyplot.show()

shift=0
pyplot.figure(figsize=(12, 6))
pyplot.plot(series[150:250], label='Measured')
pyplot.plot(pred[150-shift:250-shift], label='predicted')
pyplot.xlabel('Day')
pyplot.ylabel('Streamflow')
pyplot.title ('Predicteds (150-250)')
pyplot.legend()
pyplot.show()

# Added part #
x1=495
valid=pred[x1:x1+n_input]#series [n_train+n_test-n_input:n_train+n_test]
print (valid,n_input, n_features)
for n_phase in range(0,n_valid-5):
  x_input = valid[n_phase+0: n_phase+n_input].reshape((1, n_input, n_features)) # Bu aralık ile n_input = 3 eşit olmalı
  #print ('A',series[n_phase: n_phase+2],'B',x_input)
  yhat = model.predict(x_input, verbose=0)
  print (x_input,'--',yhat)
  valid=np.append (valid, yhat)
valid=valid[n_input:len(pred)-1]
#print('validated:',len(valid),'--',valid.shape)

pyplot.figure(figsize=(12, 6))
pyplot.plot(series[n_train+n_test-n_input:n_train+n_test-n_input+50], label='Measured')
pyplot.plot(valid[0:50], label='Valid')
pyplot.xlabel('Day')
pyplot.ylabel('Streamflow')
pyplot.title ('Validated[0:50]')
pyplot.legend()
pyplot.show()

pyplot.figure(figsize=(12, 6))
pyplot.plot(series[n_train+n_test-n_input:], label='Measured')
pyplot.plot(valid[0:2500], label='Valid')
pyplot.xlabel('Day')
pyplot.ylabel('Streamflow')
pyplot.title ('All Validated')
pyplot.legend()
pyplot.show()