@staticmethod
def sigmoid_kernel(in_gamma, r, d, x1, x2):
return math.tanh(in_gamma * np.dot(x1, x2.T) + r)
@staticmethod
def polynomial_kernel(in_gamma, r, d, x1, x2):
return math.pow(in_gamma * np.dot(x1, x2.T) + r, d)
def predict(self, features):
pre = self.intercept
for a, vec in zip(
self.alpha.reshape(self.support_vectors.shape[0], -1),
self.support_vectors):
pre += a * self.kernel(self.gamma, self.r, self.d, features, vec)
return np.asarray(pre)[0]
if __name__ == '__main__':
features, labels = BreastCancer.features, BreastCancer.label
train_feature, test_feature, train_label, test_label = ds.split(
features, labels)
regr = svm.SVR()
regr.fit(train_feature, train_label) # 训练逻辑
pred = regr.predict(test_feature) > 0.5 #预测逻辑
print(metrics.accuracy(test_label, pred))
# 手动实现向量机的前向传播计算
my_model = CustomSvr(regr)
my_pred = my_model.predict(test_feature) > 0.5
print(metrics.accuracy(test_label, pred))
class CustomComplementNB:
def __init__(self, model):
self.feature_log_pro_ = model.feature_log_prob_.T
def predict_proba(self, X):
jll = np.dot(X, self.feature_log_pro_)
log_prob_x = np.log(np.sum(np.exp(jll), axis=1))
result = jll - np.atleast_2d(log_prob_x).T
return np.exp(result)
if __name__ == '__main__':
feature, label = Iris.features, Iris.label
train_feature, test_feature, train_label, test_label = data_split.split(
feature, label)
nativeBayes = bayes.GaussianNB()
nativeBayes.fit(train_feature, train_label)
pred = nativeBayes.predict_proba(test_feature)[0]
print("pred", pred)
myBayes = CustomGaussianBayes(nativeBayes)
my_pred = myBayes.predict_proba(test_feature)[0]
print("my_pred", my_pred)
complementBayes = bayes.ComplementNB()
complementBayes.fit(train_feature, train_label)
pred = complementBayes.predict_proba(test_feature)[0]
print("pred", pred)
myComplementBayes = CustomComplementNB(complementBayes)
my_pred = myComplementBayes.predict_proba(test_feature)[0]
print("my_pred", my_pred)
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
# print(loss.numpy())
def test_step(images, labels):
predictions = model(images, training=False)
print(predictions)
predictions = np.argmax(predictions, axis=1)
print("accuracy = ", classify_metrics.accuracy(labels, predictions))
if __name__ == '__main__':
data = TextClassify1(CharSegmenter.segment)
features, labels = data.features, data.labels
train_bin_feature, test_bin_feature, train_bin_label, test_bin_label = data_split.split(features, labels)
model = MyRnnModel(data.word_dic.size() + 1)
loss_object = tf.keras.losses.SparseCategoricalCrossentropy()
optimizer = tf.keras.optimizers.Adam()
EPOCHS = 50
BATCH_SIZE = 32
for i in range(EPOCHS):
for batch in range(0, len(train_bin_feature) - BATCH_SIZE, BATCH_SIZE):
print(i, batch)
train_step(train_bin_feature[batch:batch + BATCH_SIZE][:, :100], train_bin_label[batch:batch + BATCH_SIZE])
print("-------------------------")
test_step(test_bin_feature[0:BATCH_SIZE][:, :100], test_bin_label[0:BATCH_SIZE])
for i in range(10):
result = model.custom_predict(test_bin_feature[0:BATCH_SIZE][:, :100][i])
print(result)
