Perceptron

import pandas spam_all = pandas.read_table("spambase.txt",sep=",",header=0,decimal=".")

from sklearn import metrics

spam_all.info()

from sklearn.model_selection import train_test_split X_train, X_test, Y_train, Y_test = train_test_split(spam_all.iloc[:,0:57],spam_all.loc[:,"spam"], test_size=0.2175)

from sklearn.preprocessing import StandardScaler scaler = StandardScaler(with_mean=True,with_std=True) scaler.fit(X_train) Z_train = scaler.transform(X_train) print(Z_train.mean(axis=0))

Z_test = scaler.transform(X_test) print(Z_test.mean(axis=0))

def eval_model(y_true,y_pred,poslabel="yes"): #matrice de confusion mc = metrics.confusion_matrix(y_true,y_pred) print("Confusion matrix :") print(mc) #taux d'erreur err = 1 - metrics.accuracy_score(y_true,y_pred) print("Err-rate = ",err) #F1-score f1 = metrics.f1_score(y_true,y_pred) print("F1-Score = ",f1)

from sklearn.linear_model import Perceptron ps_sklearn = Perceptron(random_state=100, max_iter = 1500, tol = None) ps_sklearn.fit(Z_train,Y_train) preds_sklearn = ps_sklearn.predict(Z_test) eval_model(Y_test,preds_sklearn) print(ps_sklearn.intercept_) coefs = ps_sklearn.coef_[0,:] print(coefs)

from sklearn.neural_network import MLPClassifier pmc_sklearn = MLPClassifier(hidden_layer_sizes=(2),activation='logistic',max_iter=1500) pmc_sklearn.fit(Z_train,Y_train) preds_sklearn = pmc_sklearn.predict(Z_test) eval_model(Y_test,preds_sklearn) print(pmc_sklearn.coefs_)

print(pmc_sklearn.intercepts_)

import numpy Z1 = numpy.zeros(Z_test.shape[0]) for i in range(Z1.shape[0]): Z1[i] = numpy.sum(Z_test[i,:]*pmc_sklearn.coefs_[0][:,0]) + pmc_sklearn.intercepts_[0][0] Z2 = numpy.zeros(Z_test.shape[0]) for i in range(Z2.shape[0]): Z2[i] = numpy.sum(Z_test[i,:]*pmc_sklearn.coefs_[0][:,1]) + pmc_sklearn.intercepts_[0][1] Z1 = 1.0/(1.0+numpy.exp(-Z1)) Z2 = 1.0/(1.0+numpy.exp(-Z2)) import matplotlib.pyplot as plt plt.scatter(Z1,Z2)

horiz = numpy.linspace(0,1) vertic = -pmc_sklearn.coefs_[1][0]/pmc_sklearn.coefs_[1][1] * horiz - pmc_sklearn.intercepts_[1]/pmc_sklearn.coefs_[1][1] plt.scatter(Z1,Z2, c=pandas.Series(['red', 'blue'])[Y_test], s=5) plt.plot(horiz, vertic, c="black", linewidth="3", linestyle="--") plt.show()

import keras print(keras.__version__)

from keras.models import Sequential from keras.layers import Dense #instanciation de la structure pmc_keras = Sequential() #architeture du réseau #entree vers couche cachée pmc_keras.add(Dense(units=2,input_dim=57,activation="sigmoid")) #couche cachée vers sortie pmc_keras.add(Dense(units=1,activation="sigmoid"))

pmc_keras.compile(loss="binary_crossentropy",optimizer="adam",metrics=["accuracy"])

pmc_keras.fit(Z_train,Y_train.astype("float"),epochs=150,batch_size=15)

proba_predm_keras = pmc_keras.predict(Z_test) #dimension print(proba_predm_keras.shape) #(1000,1) #10 premières valeurs print(proba_predm_keras[:10,:]) predm_keras = numpy.repeat(1,Z_test.shape[0]) predm_keras[proba_predm_keras[:,0] < 0.5] = 0 eval_model(Y_test,predm_keras)

print(pmc_keras.get_weights())