Tensorflow serving in Go - go

Question

I am trying to run a keras model in Go. First I train the model in python:
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(...)
Then I understand that it is advised that the model is frozen - to convert placeholder to constants.
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 = "")
When trying to use the frozen model in Golang:
model, err := tf.LoadSavedModel("frozen_my_model.pb", []string{"serve"}, nil)
It returns an error that the tag serve is not found SavedModel load for tags { serve }; Status: fail.
My questions are therefore:
How do you freeze a model in python, then load it in Go
I do this to speed up inference in Go - is it correct that freezing
models will improve inference speed?
I have noted that another function exists optimize_for_inference, how would this implemented in the above setting?

Answer 1

You have to "tag" the trained model, using
# 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()
After that, you can load it in Go.
However, a more handy solution is to use tfgo
As you can see in the README, there's the code for both: train in python and inference in Go.
I'll report here for you:
Python: train LeNet on MNIST (example)
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: inference
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)
}