def predict_pil(self, train_X, train_y, test_X, test_y, layer):
self.train_pil(train_X, train_y)
train_y_true = tools.to_categorical(train_y)
test_y_true = tools.to_categorical(test_y)
for i in range(self.pil_layers):
train_X = self.activeFunction(train_X.dot(self.pil_weight[i]), self.acFunc)
test_X = self.activeFunction(test_X.dot(self.pil_weight[i]), self.acFunc)
train_y_predict = self.get_onehot(train_X.dot(self.pil_weight[-1]))
test_y_predict = self.get_onehot(test_X.dot(self.pil_weight[-1]))
self.pil_weight = []
train_acc = accuracy_score(train_y_true, train_y_predict) * 100
test_acc = accuracy_score(test_y_true, test_y_predict) * 100
self.pil_train_acc.append(train_acc)
self.pil_test_acc.append(test_acc)
print("[INFO]+==================the {} layer pil===================".format(layer + 1))
print("PIL classifier layer {}:".format(self.pil_layers))
print("PIL Train accuracy: {} | Test accuracy: {}".format(train_acc, test_acc))
test_recall_score = recall_score(train_y_true, train_y_predict, average='micro') * 100
test_f1_score = f1_score(train_y_true, train_y_predict, average='micro') * 100
test_classification_report = classification_report(train_y_true, train_y_predict)
print("PIL test recall: {} and f1_score: {}".format(test_recall_score, test_f1_score))
# print(self.test_classification_report)
cnf_matrix = confusion_matrix(np.argmax(test_y_predict, axis=1), np.argmax(test_y_true, axis=1))
def predict_pil(self, X_train, y_train, X_test, y_test, layer_th=None):
train_X = X_train
train_y = y_train
test_X = X_test
test_y = y_test
# 调用这个predict函数 在predict之前训练
# 这样的写法有点不符合规范 但是也还说的过去
self.train_pil(train_X, train_y)
train_y_true = tools.to_categorical(train_y)
test_y_true = tools.to_categorical(test_y)
# 逐层读取保存好的权重 得到分类结果
for i in range(self.pil_layers):
train_X = self.activeFunction(train_X.dot(self.pil_weight[i]),
self.acFunc)
test_X = self.activeFunction(test_X.dot(self.pil_weight[i]),
self.acFunc)
train_loss = self.compute_loss(train_X)
test_loss = self.compute_loss(test_X)
self.train_loss.append(train_loss)
self.test_loss.append(test_loss)
train_y_predict = self.get_onehot(train_X.dot(self.pil_weight[-1]))
test_y_predict = self.get_onehot(test_X.dot(self.pil_weight[-1]))
self.pil_weight = []
train_acc = accuracy_score(train_y_true, train_y_predict) * 100
test_acc = accuracy_score(test_y_true, test_y_predict) * 100
# 将结果保存到self.pil_train_acc中
self.pil_train_acc.append(train_acc)
self.pil_test_acc.append(test_acc)
print("PIL classifier layer {}:".format(self.pil_layers))
print("PIL Train accuracy: {}% | Test accuracy: {}%".format(
train_acc, test_acc))
def random_shuffle(self, X, y):
m, n = X.shape
index = [i for i in range(m)]
random.shuffle(index)
X = X[index]
y = y[index]
y = tools.to_categorical(y)
return X, y
def train_pil(self, train_X, train_y):
X = train_X
y = tools.to_categorical(train_y)
for i in range(self.pil_layers):
pinvX = np.linalg.pinv(X)
self.pil_weight.append(pinvX)
# pinvX = pinvX[:, 0: int(self.pil_p)]
tempH = X.dot(pinvX)
X = self.activeFunction(tempH, self.acFunc)
invH = np.linalg.inv(X.T.dot(X) + np.eye(X.shape[1]) * self.pil_k) # recompute W_d
pred_W = invH.dot(X.T).dot(y)
self.pil_weight.append(pred_W)
def train_pil(self, train_X, train_y):
X = train_X
y = tools.to_categorical(train_y) # 转换成one_hot编码
# 根据self.pil_layers设置的层数 使用for循环进行逐层训练
for i in range(self.pil_layers):
pinvX = np.linalg.pinv(X)
pinvX = pinvX[:, 0:int(self.pil_p[i])]
self.pil_weight.append(pinvX)
tempH = X.dot(pinvX)
X = self.activeFunction(tempH, self.acFunc)
invH = np.linalg.inv(X.T.dot(X) + np.eye(X.shape[1]) * self.pil_k)
pred_W = invH.dot(X.T).dot(y)
self.pil_weight.append(pred_W)
def PIL_classifier(self, train_X, train_y, test_X, test_y):
train_y = tools.to_categorical(train_y)
test_y = tools.to_categorical(test_y)
self.PIL_fit(train_X, train_y)
predict_train = self.PIL_feedforward(train_X)
predict_test = self.PIL_feedforward(test_X)
train_predict = self.deal_onehot(predict_train)
test_predict = self.deal_onehot(predict_test)
self.train_acc = accuracy_score(train_predict, train_y) * 100
self.test_acc = accuracy_score(test_predict, test_y) * 100
print("==================pil=================")
print("PIL classifier layer {}:".format(self.pil_layer))
print("PIL Train accuracy: {} | Test accuracy: {}".format(
self.train_acc, self.test_acc))
self.test_recall_score = recall_score(
test_y, test_predict, average='micro') * 100
self.test_f1_score = f1_score(test_y, test_predict,
average='micro') * 100
self.test_classification_report = classification_report(
test_y, test_predict)
print("PIL test recall: {} and f1_score: {}".format(
self.test_recall_score, self.test_f1_score))
print(self.test_classification_report)