我正在尝试在 Go 中运行一个 keras 模型。首先，我在 python 中训练模型:

import keras as krs
from keras import backend as K
import tensorflow as tf

sess = tf.Session()
K.set_session(sess)
K._LEARNING_PHASE = tf.constant(0)
K.set_learning_phase(0)

m1 = krs.models.Sequential()
m1.Add(krs.layers.Dense(..., name="inputNode"))
...
m1.Add(krs.layers.Dense(..., activation="softmax", name="outputNode"))
m1.compile(...)
m1.fit(...)

然后我了解到建议卡住模型 - 将占位符转换为常量。

saver = tf.train.Saver()
tf.train.write_graph(sess.graph_def, '.', 'my_model.pbtxt')
saver.save(sess, save_path="my_model.ckpt")

from tensorflow.python.tools import freeze_graph
from tensorflow.python.tools import optimize_for_inference_lib

freeze_graph.freeze_graph(input_graph = 'my_model.pbtxt',  input_saver = "",
                 input_binary = False, input_checkpoint = "my_model.ckpt", output_node_names = "outputNode/Softmax",
                 restore_op_name = "save/restore_all", filename_tensor_name = "save/Const:0",
                 output_graph = "frozen_my_model.pb", clear_devices = True, initializer_nodes = "")

当尝试在 Golang 中使用卡住模型时:

model, err := tf.LoadSavedModel("frozen_my_model.pb", []string{"serve"}, nil)

它返回一个错误，指出没有找到标签 serve SavedModel load for tags { serve };状态:失败。

因此我的问题是:

1. 如何在 python 中卡住模型，然后在 Go 中加载它
2. 我这样做是为了加快 Go 中的推理速度 - 卡住是否正确 模型会提高推理速度？
3. 我注意到存在另一个函数 optimize_for_inference，如何在上述设置中实现它？

最佳答案

你必须“标记”训练好的模型，使用

    # Create a builder to export the model
    builder = tf.saved_model.builder.SavedModelBuilder("export")
    # Tag the model in order to be capable of restoring it specifying the tag set
    builder.add_meta_graph_and_variables(sess, ["tag"])
    builder.save()

之后，您可以在 Go 中加载它。

但是，更方便的解决方案是使用 tfgo

正如您在 README 中所见，两者都有代码:使用 Python 进行训练，使用 Go 进行推理。 我会在这里为您报告:

Python:在 MNIST 上训练 LeNet(示例)

import sys
import tensorflow as tf
from dytb.inputs.predefined.MNIST import MNIST
from dytb.models.predefined.LeNetDropout import LeNetDropout
from dytb.train import train

def main():
    """main executes the operations described in the module docstring"""
    lenet = LeNetDropout()
    mnist = MNIST()

    info = train(
        model=lenet,
        dataset=mnist,
        hyperparameters={"epochs": 2},)

    checkpoint_path = info["paths"]["best"]

    with tf.Session() as sess:
        # Define a new model, import the weights from best model trained
        # Change the input structure to use a placeholder
        images = tf.placeholder(tf.float32, shape=(None, 28, 28, 1), name="input_")
        # define in the default graph the model that uses placeholder as input
        _ = lenet.get(images, mnist.num_classes)

        # The best checkpoint path contains just one checkpoint, thus the last is the best
        saver = tf.train.Saver()
        saver.restore(sess, tf.train.latest_checkpoint(checkpoint_path))

        # Create a builder to export the model
        builder = tf.saved_model.builder.SavedModelBuilder("export")
        # Tag the model in order to be capable of restoring it specifying the tag set
        builder.add_meta_graph_and_variables(sess, ["tag"])
        builder.save()

    return 0


if __name__ == '__main__':
    sys.exit(main())

Go:推理

package main

import (
        "fmt"
        tg "github.com/galeone/tfgo"
        tf "github.com/tensorflow/tensorflow/tensorflow/go"
)

func main() {
        model := tg.LoadModel("test_models/export", []string{"tag"}, nil)

        fakeInput, _ := tf.NewTensor([1][28][28][1]float32{})
        results := model.Exec([]tf.Output{
                model.Op("LeNetDropout/softmax_linear/Identity", 0),
        }, map[tf.Output]*tf.Tensor{
                model.Op("input_", 0): fakeInput,
        })

        predictions := results[0].Value().([][]float32)
        fmt.Println(predictions)
}

关于go - Go 中的 Tensorflow 服务，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/50272602/