文章目錄

• • 1. 自定義模型
  • 2. 學(xué)習(xí)流程

學(xué)習(xí)于：簡單粗暴 TensorFlow 2

1. 自定義模型

• 重載 call() 方法，pytorch 是重載 forward() 方法

import tensorflow as tf X = tf.constant([[1.0, 2.0, 3.0],[4.0, 5.0, 6.0]]) y = tf.constant([[10.0],[20.0]])class Linear(tf.keras.Model):def __init__(self):super().__init__()self.dense = tf.keras.layers.Dense(units=1,activation=None,kernel_initializer=tf.zeros_initializer(),bias_initializer=tf.zeros_initializer())def call(self, input): # 重載 call 方法output = self.dense(input)return outputmodel = Linear()# 優(yōu)化器 optimizer = tf.keras.optimizers.SGD(learning_rate=0.001)for i in range(100):with tf.GradientTape() as tape: # 梯度記錄器y_pred = model(X)loss = tf.reduce_mean(tf.square(y_pred-y)) # 損失grads = tape.gradient(loss, model.variables) # 求導(dǎo)# 更新參數(shù)optimizer.apply_gradients(grads_and_vars=zip(grads,model.variables))

2. 學(xué)習(xí)流程

• 加載手寫數(shù)字?jǐn)?shù)據(jù)集

class MNistLoader():def __init__(self):data = tf.keras.datasets.mnist# 加載數(shù)據(jù)(self.train_data, self.train_label),(self.test_data, self.test_label) = data.load_data()# 擴展維度，灰度圖1通道 [batch_size, 28, 28, chanels=1]self.train_data = np.expand_dims(self.train_data.astype(np.float32) / 255.0, axis=-1)self.test_data = np.expand_dims(self.test_data.astype(np.float32) / 255.0, axis=-1)self.train_label = self.train_label.astype(np.int32)self.test_label = self.test_label.astype(np.int32)# 樣本個數(shù)self.num_train_data, self.num_test_data = self.train_data.shape[0], self.test_data.shape[0]def get_batch(self, batch_size):# 從訓(xùn)練集里隨機取出 batch_size 個樣本idx = np.random.randint(0, self.num_train_data, batch_size)return self.train_data[idx, :], self.train_label[idx]

• 定義模型

# 自定義多層感知機模型 class MLPmodel(tf.keras.Model):def __init__(self):super().__init__()# 除第一維以外的維度展平self.flatten = tf.keras.layers.Flatten()self.dense1 = tf.keras.layers.Dense(units=100, activation='relu')self.dense2 = tf.keras.layers.Dense(units=10)def call(self, input):x = self.flatten(input)x = self.dense1(x)x = self.dense2(x)output = tf.nn.softmax(x)return output

• 訓(xùn)練

# 參數(shù) num_epochs = 5 batch_size = 50 learning_rate = 1e-4# 模型實例 mymodel = MLPmodel() # 數(shù)據(jù)加載 data_loader = MNistLoader() # adam 優(yōu)化器 optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)num_batches = int(data_loader.num_train_data//batch_size * num_epochs) # 訓(xùn)練 for idx in range(num_batches):# 取出數(shù)據(jù)X,y = data_loader.get_batch(batch_size)with tf.GradientTape() as tape: # 梯度記錄y_pred = mymodel(X) # 預(yù)測# 計算交叉熵?fù)p失loss = tf.keras.losses.sparse_categorical_crossentropy(y_true=y, y_pred=y_pred)loss = tf.reduce_mean(loss)print("batch {}, loss {}".format(idx, loss.numpy()))# 計算梯度grads = tape.gradient(loss, mymodel.variables)# 更新參數(shù)optimizer.apply_gradients(grads_and_vars=zip(grads, mymodel.variables))

• 預(yù)測

# 評估標(biāo)準(zhǔn) sparse_categorical_accuracy = tf.keras.metrics.SparseCategoricalAccuracy() num_batches = int(data_loader.num_test_data // batch_size) # 預(yù)測 for idx in range(num_batches):# 數(shù)據(jù)區(qū)間start, end = idx*batch_size, (idx+1)*batch_size# 放入模型，預(yù)測y_pred = mymodel.predict(data_loader.test_data[start : end])# 統(tǒng)計更新 預(yù)測信息 sparse_categorical_accuracy.update_state(y_true=data_loader.test_label[start:end],y_pred=y_pred) print("test 準(zhǔn)確率：{}".format(sparse_categorical_accuracy.result())) # test 準(zhǔn)確率：0.9455000162124634

總結(jié)

以上是生活随笔為你收集整理的TensorFlow 2.0 - 自定义模型、训练过程的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。