Tensorflow.js tf.layers.activation()函数
简介: Tensorflow.js 是谷歌开发的一个开源库,用于在浏览器或节点环境中运行机器学习模型和深度学习神经网络。 Tensorflow.js tf.layers.activation()函数用于将函数应用于我们输入层的所有元素。我们还可以将函数应用于具有密集层的输入数据。
句法:
tf.layers.activation(args);
参数:以下是此函数接受的参数:
- args:它是具有字段的对象类型:
- 激活:它是应用于所有输入元素的函数的名称。
- inputShape:模型输入层的形状。它用于创建输入层。
- batchInputShape:用于输入层的制作。它为输入层中的样本定义了批次的形状。
- batchSize:用于输入层的制作。它在输入层的构造中用作batchInputShape的补充。
- dtype:定义图层的数据类型。它用于模型的第一层。
- name:它声明的字符串是输入层的名称。
- 可训练:它声明层是否可以通过函数训练。它是布尔数据类型。
- 权重:张量是层的初始数据。
- inputDType:图层中输入数据的数据类型。
返回:激活
以下是此函数的一些示例:
示例 1:在本示例中,我们将制作激活层并检查返回值。
Javascript
import * as tf from "@tensorflow/tfjs"
// Creating config for the activation layer
const config = {
activation: 'sigmoid',
inpurShape: 5,
dtype: 'int32',
name: 'activationLayer'
};
// Defining the activation layer
const activationLayer = tf.layers.activation(config);
// printing return of activation layer
console.log(activationLayer);
Javascript
import * as tf from "@tensorflow/tfjs"
// Configuration file for the activation layer
const geek_config = {
activation: 'sigmoid',
inpurShape: 5,
dtype: 'int32',
name: 'activationLayer'
};
const geek_activation = tf.layers.activation(geek_config);
const geek_inputLayer = tf.layers.dense({units: 1});
// Our Input layer for the model
const geek_input = tf.input({shape: [7]});
// Making structure for the model
const geek_output = geek_inputLayer.apply(geek_input);
const geek_result = geek_activation.apply(geek_output);
// Making Model from structure
const config2 = {inputs: geek_input, outputs: geek_result}
const model = tf.model(config2);
// Collect both outputs and print separately.
const config3 = tf.randomUniform([4, 7])
const geek_activationResult = model.predict(confg3);
geek_activationResult.print();
输出:
{
"_callHook": null,
"_addedWeightNames": [],
"_stateful": false,
"id": 38,
"activityRegularizer": null,
"inputSpec": null,
"supportsMasking": true,
"_trainableWeights": [],
"_nonTrainableWeights": [],
"_losses": [],
"_updates": [],
"_built": false,
"inboundNodes": [],
"outboundNodes": [],
"name": "ActivationLayer",
"trainable_": true,
"initialWeights": null,
"_refCount": null,
"fastWeightInitDuringBuild": false,
"activation": {}
}
示例 2:在此示例中,我们将使用一些配置创建我们的激活层,并使用激活层训练我们的输入数据。
Javascript
import * as tf from "@tensorflow/tfjs"
// Configuration file for the activation layer
const geek_config = {
activation: 'sigmoid',
inpurShape: 5,
dtype: 'int32',
name: 'activationLayer'
};
const geek_activation = tf.layers.activation(geek_config);
const geek_inputLayer = tf.layers.dense({units: 1});
// Our Input layer for the model
const geek_input = tf.input({shape: [7]});
// Making structure for the model
const geek_output = geek_inputLayer.apply(geek_input);
const geek_result = geek_activation.apply(geek_output);
// Making Model from structure
const config2 = {inputs: geek_input, outputs: geek_result}
const model = tf.model(config2);
// Collect both outputs and print separately.
const config3 = tf.randomUniform([4, 7])
const geek_activationResult = model.predict(confg3);
geek_activationResult.print();
输出:
Tensor
[[0.4178988],
[0.2027801],
[0.2813435],
[0.2546847]]
参考: https://js.tensorflow.org/api/latest/#layers.activation