def main():
if len(sys.argv) < 6:
error = "\n Error: The number of parameter is wrong\
\n\n\t python Classifier.py model_type validation\
\n\n\t model_type: 1 to 6 and 7 for all model\
\n\t cross_validation: 0-1\
\n\t Evaluation of the model : 0-1\
\n\t Performance of the model(Learning Curve) : 0-1\
\n\t Bagging: 0-1\n"
print(error)
return
type_model = int(sys.argv[1])
cross_val = bool(int(sys.argv[2]))
eval_model = bool(int(sys.argv[3]))
perf_model = bool(int(sys.argv[4]))
bagging = bool(int(sys.argv[5]))
# Load database
print("Loading Database")
dm = DM.DataManager(200, 100, normalisation=False)
x_train, t_train, x_test, t_test = dm.generer_donnees()
# Deep layer
num_layer = 20
num_neural_per_layer = 400
layer_sizes = (num_neural_per_layer, ) * num_layer
Specific_tuple = (200, )
Specific_tuple2 = (200, 400, 600, 620)
# Bagging
number_model = 50
md = None
title = None
# Selection of the model
if type_model == 1:
print("Selecting Perceptron")
title = "Model 1"
if bagging:
md = Model.Bagging(base_model='Perceptron',
number_model=number_model,
reg_penalty='l2',
reg=0.001)
else:
md = Model.Perceptron(reg_penalty='l2', reg=0.001, k_fold=5)
elif type_model == 2:
print("Selecting MLPerceptron")
title = "Model 2"
if bagging:
md = Model.Bagging(base_model='MLPerceptron',
number_model=number_model,
hidden_layer_sizes=Specific_tuple,
activation='relu',
reg=0.001)
else:
md = Model.MLPerceptron(hidden_layer_sizes=Specific_tuple,
activation='relu',
reg=0.001,
k_fold=5)
elif type_model == 3:
print("Selecting Decision Tree")
title = "Model 3"
if bagging:
md = Model.Bagging(base_model='ModelDecisionTree',
number_model=number_model,
criterion='gini')
else:
md = Model.ModelDecisionTree(criterion='gini')
elif type_model == 4:
print("Selecting SVM")
title = "Model 4"
if bagging:
md = Model.Bagging(base_model='ModelDecisionTree',
number_model=number_model,
kernel="poly",
degree=2)
else:
md = Model.ModelSVM(kernel="poly", degree=2, verbose=True)
elif type_model == 5:
print("Selecting Logistic Regression")
title = "Model 5"
if bagging:
md = Model.Bagging(base_model='LogisticRegression',
number_model=number_model,
reg_penalty='l2',
reg=1.0)
else:
md = Model.LogisticRegression(reg_penalty='l2',
reg_inv=1.0,
k_fold=5)
else:
print("Error : No model is train.")
# Training of the model
if cross_val is False:
md.train(x_train, t_train)
else:
md.cross_validation(x_train, t_train)
# Compute Train / Test
# Prediction on the train dataset and the test dataset
predictions_entrainement = np.array([md.prediction(x) for x in x_train])
train_error = np.sum([md.error(x, t) for (x, t) in zip(x_train, t_train)])
print("Train error = ", 100 * train_error / len(t_train), "%")
predictions_test = np.array([md.prediction(x) for x in x_test])
test_error = np.sum([md.error(x, t) for (x, t) in zip(x_test, t_test)])
print("Test error = ", 100 * test_error / len(t_test), "%")
analyzer = Utils.ModelAnalyzer()
if eval_model is True:
print(analyzer.confusionMatrix(t_test, predictions_test))
print("The accuracy is : ", analyzer.accuracy(t_test,
predictions_test))
x = np.append(x_train, x_test, axis=0)
t = np.append(t_train, t_test)
if perf_model is True: # Change parameter for the exemple
plt = analyzer.plotLearningCurves(md.mod, x, t, title=title)
plt.show()
