print("Classification Number:", (c+1))
# Generating training and testing set
x_train = np.empty((0,feat))
x_test = np.empty((0,feat))
for genere in genres.keys():
np.random.shuffle(genres[genere])
x_train = np.append(x_train, (genres[genere])[0:75, :] , axis = 0)
x_test = np.append(x_test, (genres[genere]) [75:100, :] , axis = 0)
# predictions for each classifier
# SVM improves with KNN
svm_predictions = classifiers.svm(x_train, y_train, x_test)
# Adaptive Boosting
ada_predictions = classifiers.boost(x_train, y_train, x_test)
# Naive Baesian
bay_predictions = classifiers.bayes(x_train, y_train, x_test)
# Choose final prediction with rules:
# 1. If baesian states one between "Classical", "Jazz", "Metal", "Reggae" pick it
# 2. else if every classifier or 2/3 state the same pick that choice
# 3. If every classifier states something different focus on Boost and SVM:
# if both state on of: "Disco", "Jazz", "Metal", "Pop" or "Reggae" pick SVM
# else if both state one of others pick Boost
# else pick random between the two
final_prediction = []
random.shuffle(train_set)
Y_train = np.array([row[0] for row in train_set])
X_train = np.array([row[1] for row in train_set])
test_set = []
for key, value in test.items():
for file_id, words in value.items():
bow = np.zeros(len(vocab))
for word in words:
if word in vocab:
bow[vocab[word]] += 1
else:
bow[vocab['</unk>']] += 1
tf = bow/len(words)
tfidf = tf*idf_test
avg_glove = np.zeros(300)
for i in range(len(vocab)):
avg_glove += tfidf[i]*glove[i]
test_set.append([key, avg_glove])
print('shuffling test')
random.shuffle(test_set)
Y_test = np.array([row[0] for row in test_set])
X_test = np.array([row[1] for row in test_set])
print("Calling Classifiers\n")
print("Accuracy for Naive Bayes is : ", naive_bayes(X_train, Y_train, X_test, Y_test))
print("Accuracy for Logistic Regression is : ", logistic_regression(X_train, Y_train, X_test, Y_test))
print("Accuracy for SVM is : ", svm(X_train, Y_train, X_test, Y_test))
print("Accuracy for FF Neural Net is : ", fnn(X_train, Y_train, X_test, Y_test))
# print("Accuracy for Recurrent Neural Net is : ", rnn(X_train, Y_train, X_test, Y_t))
print('Calculando Melhores Parametros\n')
bp_dt = classifiers.fbp_dt()
params_dt.append(bp_dt)
bp_knn = classifiers.fbp_knn()
params_knn.append(bp_knn)
bp_mlp = classifiers.fbp_mlp()
params_mlp.append(bp_mlp)
print('Naive Bayes\nUsando Configuração Padrão')
bp_svm = classifiers.fbp_svm()
params_svm.append(bp_svm)
dt.append(classifiers.dt(bp_dt))
knn.append(classifiers.knn(bp_knn))
mlp.append(classifiers.mlp(bp_mlp))
nb.append(classifiers.nb())
svm.append(classifiers.svm(bp_svm))
with open("results.csv", "w") as fp:
fp.write(
"Dt, Depth, Knn, Weights, K, Mlp, Max Ite, Hidden Layers, Learning Rate, Nb, No Params, Svm, Kernel, Penalty\n"
)
for index in range(20):
fp.write(
"%f, %s, %f, %s, %s, %f, %s, %s, %s, %f, %s, %f, %s,%s\n" %
(dt[index], params_dt[index], knn[index],
'1-Distancia Normalizada' if inspect.isfunction(
params_knn[index][0]) else params_knn[index][0],
params_knn[index][1], mlp[index], params_mlp[index][0],
params_mlp[index][1], params_mlp[index][2], nb[index], "Default",
svm[index], params_svm[index][0], params_svm[index][1]))
fp.write("\n%f, %s, %f, %s, %s, %f, %s, %s, %s, %f, %s, %f, %s,%s\n" %