Skip to content

Commit

Permalink
20210508
Browse files Browse the repository at this point in the history
  • Loading branch information
@sun1638650145
sun1638650145 committed May 8, 2022
1 parent e11c3a5 commit cd4d6b9
Showing 1 changed file with 155 additions and 30 deletions.
185 changes: 155 additions & 30 deletions TensorFlow.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1490,7 +1490,17 @@ from tensorflow.keras import layers
layer = layers.GlobalMaxPooling1D()
```

#### 1.14.7.19.GRU()
#### 1.14.7.19.GlobalMaxPooling2D()

实例化全局2D最大池化层.

```python
from tensorflow.keras import layers

layer = layers.GlobalMaxPooling2D()
```

#### 1.14.7.20.GRU()

实例化门控循环网络层.

Expand All @@ -1501,7 +1511,7 @@ layer = layers.GRU(units=256, # int|神经元的数量.
return_sequences=True) # bool|False|是否返回全部序列.
```

#### 1.14.7.20.Input()
#### 1.14.7.21.Input()

实例化输入层.

Expand All @@ -1513,7 +1523,7 @@ layer = layers.Input(shape=(224, 224, 3), # tuple|输入张量的形状.
dtype=None) # str|None|期望的数据类型.
```

#### 1.14.7.21.InputLayer()
#### 1.14.7.22.InputLayer()

实例化输入层.

Expand All @@ -1523,7 +1533,7 @@ from tensorflow.keras import layers
layer = layers.InputLayer(input_shape=(224, 224, 3)) # tuple|输入张量的形状.
```

#### 1.14.7.22.Lambda()
#### 1.14.7.23.Lambda()

将一个函数封装称网络层.

Expand All @@ -1535,7 +1545,7 @@ layer = layers.Lambda(function=lambda x: x**2, # lambda or function|要封装
name=None) # str|None|网络层的名称.
```

#### 1.14.7.23.Layer()
#### 1.14.7.24.Layer()

自定义一个符合tf.keras接口的层.

Expand All @@ -1552,7 +1562,7 @@ class MyLayer(layers.Layer):
return outputs
```

##### 1.14.7.23.1.get_weights()
##### 1.14.7.24.1.get_weights()

获取当前网络层的权重.|list of `numpy.ndarray`

Expand All @@ -1565,23 +1575,23 @@ layer(tf.constant([[1., 2., 3.]]))
weights = layer.get_weights()
```

##### 1.14.7.23.2.output
##### 1.14.7.24.2.output

网络层输出张量.|`tensorflow.python.keras.engine.keras_tensor.KerasTensor`

```python
tensor = layer.output
```

##### 1.14.7.23.3.trainable
##### 1.14.7.24.3.trainable

网络层在训练期间是否能更新权重.|`bool`

```python
trainable = layer.trainable
```

#### 1.14.7.24.LeakyReLU()
#### 1.14.7.25.LeakyReLU()

实例化带泄漏的ReLU层.

Expand All @@ -1591,7 +1601,7 @@ from tensorflow.keras import layers
layer = layers.LeakyReLU(alpha=0.3) # float|0.3|负斜率系数(泄漏率).
```

#### 1.14.7.25.LSTM()
#### 1.14.7.26.LSTM()

实例化长短时记忆层.

Expand All @@ -1603,7 +1613,7 @@ layer = layers.LSTM(units=256, # int|神经元的数量.
dropout=0.) # float|0.|随机丢弃比例.
```

#### 1.14.7.26.MaxPooling1D()
#### 1.14.7.27.MaxPooling1D()

实例化1D最大池化层.

Expand All @@ -1615,7 +1625,7 @@ layer = layers.MaxPooling1D(pool_size=2, # int|2|池化窗口.
padding='valid') # {'valid', 'same'}|'valid'|填充方式.
```

#### 1.14.7.27.MaxPooling2D()
#### 1.14.7.28.MaxPooling2D()

实例化2D最大池化层.

Expand All @@ -1627,7 +1637,7 @@ layer = layers.MaxPooling2D(pool_size=(2, 2), # int or tuple of 2 int|(2, 2)|
padding='valid') # {'valid', 'same'}|'valid'|填充方式.
```

#### 1.14.7.28.ReLU()
#### 1.14.7.29.ReLU()

实例化ReLU层.

Expand All @@ -1637,7 +1647,7 @@ from tensorflow.keras import layers
layer = layers.ReLU()
```

#### 1.14.7.29.Reshape()
#### 1.14.7.30.Reshape()

实例化变形层.

Expand All @@ -1647,7 +1657,7 @@ from tensorflow.keras import layers
layer = layers.Reshape(target_shape=(None, 10)) # tuple of int|目标形状.
```

#### 1.14.7.30.SeparableConv2D()
#### 1.14.7.31.SeparableConv2D()

实例化深度可分离2D卷积层.

Expand All @@ -1660,7 +1670,7 @@ layer = layers.SeparableConv2D(filters=32, # int|卷积核的数量.
padding='valid') # {'valid', 'same' or 'causal'}|'valid'|填充方式.
```

#### 1.14.7.31.SimpleRNN()
#### 1.14.7.32.SimpleRNN()

实例化循环网络层.

Expand All @@ -1672,7 +1682,7 @@ layer = layers.SimpleRNN(units=256, # int|神经元的数量.
return_sequences=True) # bool|False|是否返回全部序列.
```

#### 1.14.7.32.TimeDistributed()
#### 1.14.7.33.TimeDistributed()

实例化时间片封装器.

Expand All @@ -1683,7 +1693,7 @@ layer = layers.Dense(32)
layer = layers.TimeDistributed(layer=layer) # keras.layers|需要分片的网络层.
```

#### 1.14.7.33.UpSampling2D()
#### 1.14.7.34.UpSampling2D()

```python
from tensorflow.keras import layers
Expand All @@ -1692,7 +1702,7 @@ from tensorflow.keras import layers
layer = layers.UpSampling2D(size=(2, 2)) # int or tuple of 2 int|(2, 2)|上采样因子.
```

#### 1.14.7.34.ZeroPadding2D()
#### 1.14.7.35.ZeroPadding2D()

实例化2D零填充层.

Expand Down Expand Up @@ -1769,7 +1779,18 @@ from tensorflow.keras.losses import MeanAbsoluteError
loss = MeanAbsoluteError()
```

#### 1.14.8.6.SparseCategoricalCrossentropy()
#### 1.14.8.6.mean_squared_error()

计算均方误差值.|`tensorflow.python.framework.ops.EagerTensor`

```python
from tensorflow.keras.losses import mean_squared_error

value = mean_squared_error(y_true=[1, 2, 3, 4], # array-like|真实值.
y_pred=[2, 2, 3, 4]) # array-like|预测值.
```

#### 1.14.8.7.SparseCategoricalCrossentropy()

实例化多分类交叉熵损失函数(稀释编码).

Expand All @@ -1787,14 +1808,44 @@ loss = SparseCategoricalCrossentropy(from_logits=False) # bool|False|是否将

#### 1.14.9.1.MAE()

实例化平均绝对误差损失函数.
实例化平均绝对误差评估函数.

```python
from tensorflow.keras.metrics import MAE

metric = MAE()
```

#### 1.14.9.2.Mean()

实例化计算均值评估函数.

```python
from tensorflow.keras.metrics import Mean

metric = Mean()
```

##### 1.14.9.2.1.result()

计算评估值.|`tensorflow.python.framework.ops.EagerTensor`

```python
from tensorflow.keras.metrics import Mean

metric = Mean()
metric.update_state([1, 2, 3])
value = metric.result()
```

##### 1.14.9.2.2.update_state()

累积用于计算评估的值.

```
metric.update_state(values=[1, 2, 3]) # array-like|要统计的值.
```

### 1.14.10.mixed_precision

| 版本 | 描述 | 注意 |
Expand Down Expand Up @@ -1988,14 +2039,15 @@ layers = model.layers
model.load_weights(filepath) # str or pathlib.Path|文件路径.
```

##### 1.14.11.2.9.predict()
##### 1.14.11.2.9.metrics()

使用模型进行预测.|`numpy.ndarray`
返回向模型添加的评估函数.

```python
y_pred = model.predict(x, # Numpy array, TensorFlow tensor, `tf.data` dataset, generator or `keras.utils.Sequence`|None|特征数据.
batch_size=None, # int|None|批次大小.
verbose=0) # int|0|日志显示模式.
@property
def metrics(self):
"""列出全部的评估函数, 使得每个epoch后评估函数将会自动重置."""
return [metric_0, metric_1]
```

##### 1.14.11.2.10.output_shape
Expand All @@ -2006,7 +2058,17 @@ y_pred = model.predict(x, # Numpy array, TensorFlow tensor, `tf.data` dataset,
shape = model.output_shape
```

##### 1.14.11.2.11.save()
##### 1.14.11.2.11.predict()

使用模型进行预测.|`numpy.ndarray`

```python
y_pred = model.predict(x, # Numpy array, TensorFlow tensor, `tf.data` dataset, generator or `keras.utils.Sequence`|None|特征数据.
batch_size=None, # int|None|批次大小.
verbose=0) # int|0|日志显示模式.
```

##### 1.14.11.2.12.save()

保存模型.

Expand All @@ -2015,22 +2077,74 @@ model.save(filepath='./model.h5', # str or pathlib.Path|文件路径.
save_format=None) # {'tf', 'h5'}|None|保存文件格式.
```

##### 1.14.11.2.12.summary()
##### 1.14.11.2.13.summary()

打印模型的摘要.

```python
model.summary()
```

##### 1.14.11.2.13.trainable_variables
##### 1.14.11.2.14.test_step()

实现自定义评估步骤函数.

```python
def test_step(self, data):
"""使用自定义的评估步骤, 即可继续使用model.evaluate()."""
if len(data) == 3:
x, y, sample_weight = data # 提取样本权重.
else:
sample_weight = None
x, y = data

# 前向传播.
y_pred = self(x, training=False)
# 计算损失.
self.compiled_loss(y, y_pred, sample_weight=sample_weight)
# 更新评估值.
self.compiled_metrics.update_state(y, y_pred, sample_weight=sample_weight)

return {m.name: m.result() for m in self.metrics} # 返回评估值字典.
```

##### 1.14.11.2.15.trainable_variables

模型所有可训练权重的列表.|`list`

```python
model.trainable_variables
```

##### 1.14.11.2.16.train_step()

实现自定义训练步骤函数.

```python
def train_step(self, data):
"""使用自定义的训练步骤, 即可继续使用model.fit()."""
if len(data) == 3:
x, y, sample_weight = data # 提取样本权重.
else:
sample_weight = None
x, y = data

with tf.GradientTape() as tape:
# 前向传播.
y_pred = self(x, training=True)
# 计算损失.
loss = self.compiled_loss(y, y_pred, sample_weight=sample_weight)

# 反向传播, 计算梯度.
gradients = tape.gradient(loss, self.trainable_variables)
# 更新参数.
self.optimizer.apply_gradients(zip(gradients, self.trainable_variables))
# 更新评估值.
self.compiled_metrics.update_state(y, y_pred, sample_weight=sample_weight)

return {m.name: m.result() for m in self.metrics} # 返回评估值字典.
```

#### 1.14.11.3.Sequential()

实例化`Sequential`.
Expand Down Expand Up @@ -2789,7 +2903,18 @@ tensor = tf.where(condition=(a > b), # tf.Tensor of type bool|判断条件.
y=False) # tf.Tensor|None|情况为假的返回值.
```

## 1.38.zeros_like()
## 1.38.zeros()

创建全零张量.|`tensorflow.python.framework.ops.EagerTensor`

```python
import tensorflow as tf

tensor = tf.zeros(shape=(3, 2), # list/tuple of int|张量的形状.
dtype='int32') # str|dtypes.float32|元素数据类型.
```

## 1.39.zeros_like()

创建一个输入数组形状相同的全零张量.|`tensorflow.python.framework.ops.EagerTensor`

Expand Down

0 comments on commit cd4d6b9

Please sign in to comment.