TP-APP-SUPERVISE/TP2/MLP.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Dec 15 18:26:31 2021

@author: chouiya
"""


#********import***************
from sklearn.datasets import fetch_openml
from sklearn import datasets 
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
import numpy as np
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import precision_score
from sklearn.metrics import zero_one_loss
import time
#*******************MLP pour une seule couche de 50 neurons*****************
mnist = fetch_openml('mnist_784',as_frame=False)

data=mnist.data
target=mnist.target
xtrain, xtest, ytrain, ytest = train_test_split(data, target, train_size=0.7)

clf = MLPClassifier(hidden_layer_sizes=(50))

clf.fit(xtrain, ytrain)
prediction = clf.predict(xtest)
score = clf.score(xtest, ytest)
precision = precision_score(ytest, prediction, average ='micro')
loss1_0 = zero_one_loss(ytest, prediction)

# Print & Test :------
print("This MLP model, with one layer of 50, has a score of : ", score*100, "%.")
print(" 4 th image : Prediction ",prediction[3], "Vs  Reel : ", ytest[3])

# Showing the 4th image:
images = xtest.reshape((-1, 28, 28))
plt.imshow(images[3],cmap=plt.cm.gray_r,interpolation="nearest")
plt.show()

# Metrics :
print ("This MLP model has a precision of :", precision*100, "%.")
print ("This MLP model has a zero-one_loss of :",loss1_0*100, "%.")

#*******************Variation de nombre de couche de 2 à 100*******
hidden_layer =(50,)*100

Scor = []
Pred= []
Loss= []

for i in range (100):
    clf = MLPClassifier(hidden_layer_sizes = hidden_layer[0:i])
    clf.fit(xtrain, ytrain)
    prediction = clf.predict(xtest)
    score = clf.score(xtest, ytest)
    precision = precision_score(ytest, prediction, average='micro')
    loss0_1 = zero_one_loss(ytest, prediction)

    Scor.append(score)
    Pred.append(precision)
    Loss.append(loss0_1)

    print("For ", i, "hidden layer (s), The score = ", score *100, "%", ", Precision = ", precision*100, "% .." )


fig, ax = plt.subplots(3, sharex=True, figsize=(10,10))
ax[0].plot(range(100), Scor)
ax[0].set_title('Number of hidden layers from 1 to 99')
ax[0].set_ylabel('Score')
ax[1].plot(range(100), Pred)
ax[1].set_ylabel('Precision')
ax[2].plot(range(100), Loss)
ax[2].set_ylabel('Zero-to-one Loss')

#**************les 5 modeles de classification*********
clf1 = MLPClassifier(hidden_layer_sizes=(300))
# 2 layers
clf2 = MLPClassifier(hidden_layer_sizes=(20, 50))
# 4 layers
clf4 = MLPClassifier(hidden_layer_sizes=(20,50, 100, 150))
# 6 layers
clf6 = MLPClassifier(hidden_layer_sizes=( 20, 50, 150, 100, 50, 10))
# 8 layers, increase neurals :
clf8 = MLPClassifier(hidden_layer_sizes=(20, 40, 60, 120, 150, 180, 200, 250))

ClassifierList = ("clf1", "clf2","clf4", "clf6", "clf8")

Score =[]
Precision = []
Loss = []
TimeTraining = []
TimePrediction = []

def clfs(clf, i):

    #Training :
    startTrain =time.time()
    clf.fit(xtrain, ytrain)
    endTrain = time.time()

    #Prediction :
    startpred= time.time()
    predict = clf.predict(xtest)
    endpred = time.time()

    #Metrics :
    score = clf.score(xtest,ytest)
    precision =  precision_score(ytest, predict,  average='micro')
    loss01 = zero_one_loss(ytest, predict)
    timetrain = endTrain - startTrain
    timePred = endpred - startpred

    #Saving results
    Score.append(score*100)
    Precision.append(precision*100)
    Loss.append(loss01)
    TimePrediction.append(timePred)
    TimeTraining.append(timetrain)

    #Prints :
    print("For the", i," model we have, score = ", score*100, "%, precision =",precision*100, "%." )
    print("   Training's time = ", timetrain, " and prediction's time = ", timePred, "." )

    
#***********************plot*******
fig, ax = plt.subplots(5, sharex=True, figsize=(10,10))
ax[0].scatter(range(5), Score, c='orange')
ax[0].set_title('The five classifiers with 1,2,4,6,8 hidden layers')
ax[0].set_ylabel('Score (%)')
ax[1].scatter(range(5), Precision, c='red')
ax[1].set_ylabel('Precision (%)')
ax[2].scatter(range(5), Loss, c='blue')
ax[2].set_ylabel('Zero-to-one Loss')
ax[3].scatter(range(5), TimeTraining, c='pink')
ax[3].set_ylabel('Training Time (s)')
ax[4].scatter(range(5), TimePrediction, c='purple')
ax[4].set_ylabel('¨Prediction Time (s)')

plt.show()

#*****Etude de la convergence des algorithmes d'optimisation : adam, sgd, lbfgs***

tab1=['adam','sgd','lbfgs']
tab2=['relu','tanh','logistic','identity']
for i in tab1:# solver
  for j in tab2:
    #activation function
    clf = MLPClassifier(hidden_layer_sizes =50,activation=j,solver=i,verbose=False)
    clf.fit(xtrain, ytrain)
    prediction = clf.predict(xtest)
    score = clf.score(xtest, ytest)
    precision = precision_score(ytest, prediction, average='micro')
    loss0_1 = zero_one_loss(ytest, prediction)
    print('the result of the solver',i,'and the activation function',j)
    Scor.append(score)
    Pred.append(precision)
    Loss.append(loss0_1)


print('score :',Scor)
print('prediction',Pred)
print('loss',Loss)

#**********Variation de alpha**************
alphas = np.logspace(-5, 3, 5)
 for i in alphas:
   
    clf = MLPClassifier(hidden_layer_sizes =50,activation='relu',solver='adam',alpha=i,verbose=False)
    clf.fit(xtrain, ytrain)
    prediction = clf.predict(xtest)
    score = clf.score(xtest, ytest)
    precision = precision_score(ytest, prediction, average='micro')
    loss0_1 = zero_one_loss(ytest, prediction)
    print('for alpha equal to: ',i)
    Scor.append(score)
    Pred.append(precision)
    Loss.append(loss0_1)
    
print('score :',Scor)
print('prediction',Pred)
print('loss',Loss)
#***************










TP2_CNN.py
Affichage de TP2_CNN.py en cours...