Train_Algorithm.py

import BuildMatrices
from sklearn import cross_validation
from sklearn.ensemble import GradientBoostingRegressor
import numpy as np
import pickle
import warnings
import copy

### This script is responsible for calling the data extraction script and matrix builder script in order to create the models for each stage.
### The models are then pickled so that they can easily be accessed later on
                
# load the dictionary of sparse matrices as well as the column dictionary
dat,col = BuildMatrices.build_matrices('devAEHRA')
print('\n')
print('***************************************************************')
print('*                   Creating and Saving Models                *')
print('***************************************************************\n')

#Loop through the dictionary of sparse matrices to create and save models so that they can be called when a prediction will be needed
for key in dat: 
    if key == 'RFT': #We are not interested in the Ready For Treatment Stage so we discard that
        continue
    print('Building Model for ' + key + '... ', end='') 
    X = dat[key][:,0:-1]
    y = dat[key][:,-1]


    # Create model for each stage. I used the Gradient Bossting Regressor algorithm since it gave me the best results.
    model  = GradientBoostingRegressor(loss = 'lad',n_estimators=100).fit(X,y)
    
    # save each model in a encoded pickle file named by the key
    file = open('/var/www/devDocuments/marc/ML_Algorithm_MUHC/'+key+'.pkl','wb')
    pickle.dump(model,file)
    file.close()
    # save the column dictionary (dictionary that has all the features for each key)
    file = open('/var/www/devDocuments/marc/ML_Algorithm_MUHC/columndictionary.pkl','wb')
    pickle.dump(col,file)
    file.close()
    
    print('Done')

    # The following code is for observational purposes only. It gives a rough approximation of the performance for each model by returning
    # the absolute mean error. This is useful for research purposes and hence is disactivated for server training purposes.
    '''
    scores = cross_validation.cross_val_score(model, X, y, cv=10, scoring = 'mean_absolute_error')
    mean_abs_err = np.abs(np.mean(scores))
    print('Done Gradient Boosting Regressor. Mean absolute error: %.3f \n' %(mean_abs_err))
    '''

print('\nThe algorithm has been successfully trained')
print('use the "predictor.py" script to make predictions')
    

# A little function forprinting purposes. Gives the console a better look.
def print_status(done,key):
    if done:
        print('Building Model for ' + key + '... Done')
    else:
        print('Building Model for ' + key + '...')