#! /usr/bin/env python3
#
def hourly_wages ( ):

#*****************************************************************************80
#
## hourly_wages uses keras to solve a multivariate regression case.
#
#  Licensing:
#
#    This code is distributed under the MIT license.
#
#  Modified:
#
#    29 April 2019
#
#  Author:
#
#    John Burkardt.
#
  import keras
  import matplotlib.pyplot as plt
  import numpy as np
  import platform
  
  print ( '' )
  print ( 'hourly_wages:' )
  print ( '  python version: %s' % ( platform.python_version ( ) ) )
  print ( '  keras version: %s' % ( keras.__version__ ) )
  print ( '  Neural network to solve a multivariate regression problem.' )
  print ( '  The data is read from an external file.' )
  print ( '' )
#
#  Load the data.
#
  data = np.loadtxt ( 'hourly_wages.csv', delimiter = ',', skiprows = 1 )
  print ( '  Data contains %d records with %d features.' % ( data.shape[0], data.shape[1] ) )
#
#  Data column 1 is the wage, which will be our target.
#  Data columns 2-10 are features.
#  Extract 450 records as training, 84 as test.
#
  train_data = data[0:450,1:10]
  train_targets = data[0:450,0]
  train_num = train_data.shape[0]
  feature_num = train_data.shape[1]
  print ( '  Training data uses %d records with %d features and 1 target.' % ( train_num, feature_num ) )

  test_data = data[450:534,1:10]
  test_targets = data[450:534,0]
  test_num = test_data.shape[0]
  print ( '  Test data uses %d records with %d features and 1 target.' % ( test_num, feature_num) )
#
#  Exhibit the contents of one item of the training data:
#
  print ( '  train_data[0,0:9]:' )
  print ( train_data[0,0:9] )
  print ( '  train_targets[0]:' )
  print ( train_targets[0] )
#
#  Build the model.
#
  from keras import models
  from keras import layers

  model = models.Sequential()
  model.add ( layers.Dense ( 15, activation = 'relu', input_shape = ( feature_num, ) ) )
  model.add ( layers.Dense ( 15, activation = 'relu' ) )
  model.add ( layers.Dense ( 1 ) )
#
#  Compile the model.
#
  model.compile ( optimizer = 'adam', loss = 'mean_squared_error' )
#
#  Allow for early stopping if training is not improving the estimate.
#
  from keras.callbacks import EarlyStopping

  early_stopping_monitor = EarlyStopping ( patience = 3 )
#
#  Train the model.
#
  print ( '' )
  print ( 'Training:' )
  print ( '' )

  model.fit ( train_data, train_targets, validation_split = 0.2, epochs = 30, \
    callbacks = [ early_stopping_monitor ] )
#
#  Test
#
  print ( '' )
  print ( 'Testing:' )
  print ( '' )
  test_predictions = model.predict ( test_data )
  for i in range ( test_num ):
    print ( '  %d: %f  %f' % ( i, test_predictions[i], test_targets[i] ) )
#
#  Terminate.
#
  print ( '' )
  print ( 'hourly_wages:' )
  print ( '  Normal end of execution.' )
  return

def timestamp ( ):

#*****************************************************************************80
#
## TIMESTAMP prints the date as a timestamp.
#
#  Licensing:
#
#    This code is distributed under the MIT license. 
#
#  Modified:
#
#    06 April 2013
#
#  Author:
#
#    John Burkardt
#
#  Parameters:
#
#    None
#
  import time

  t = time.time ( )
  print ( time.ctime ( t ) )

  return None

if ( __name__ == '__main__' ):
  timestamp ( )
  hourly_wages ( )
  timestamp ( )