如何在 tensorflow.js 中使用现有的 keras 模型

Question

我有 Keras 模型，已转换为 tensorflow.js，但无法在 javascript 中加载模型，具体步骤是什么？

model.add(Embedding(vocabulary_size, seq_len, input_length=seq_len))
model.add(LSTM(256,return_sequences=True))
model.add(LSTM(128))
model.add(Dense(256,activation='relu'))
model.add(Dense(vocabulary_size, activation='softmax'))
# compiling the network
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(train_inputs,train_targets,epochs=256,verbose=1)

Answer 1

import tensorflowjs as tfjs

model.add(Embedding(vocabulary_size, seq_len, input_length=seq_len))
model.add(LSTM(256,return_sequences=True))
model.add(LSTM(128))
model.add(Dense(256,activation='relu'))
model.add(Dense(vocabulary_size, activation='softmax'))
# compiling the network
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(train_inputs,train_targets,epochs=256,verbose=1)

# save the model in model.json

tfjs.converters.save_keras_model(model, './keras_converted')

在 javascript 中加载模型

• 要在浏览器中加载 model.json，它必须在本地服务器中提供，因为浏览器无法直接访问文件系统

import * as tf from '@tensorflow/tfjs';

const model = await tf.loadLayersModel('https://hostname:port/path/to/model.json');

Answer 2

您可以像这样从端点加载模型。

import * as tf from '@tensorflow/tfjs';
import React, {useState, useEffect} from "react";

const url = {
    model: 'http://localhost:81/tfjs-models/model.json',
};

async function loadModel(url) {
  try {
    // For layered model
    const model = await tf.loadLayersModel(url.model);
    // For graph model
    // const model = await tf.loadGraphModel(url.model);
    setModel(model);
    console.log("Load model success");
  } catch (err) {
    console.log(err);
  }
} 
const [model, setModel] = useState();

useEffect(() => {
  tf.ready().then(() => {
    loadModel(url);
  });
}, []);

然后可以使用model from state 访问模型。