how to send two placeholder varialbles with remote execute

[wangce888]

x = tf.placeholder(tf.int32, shape=[1], name='x') y = tf.placeholder(tf.int32, shape=[1], name='y') b = tf.Variable(1, name='b') xy = tf.multiply(x, y, name='multiply') output = tf.add(xy, b, name='op_to_store')

with tf.Session() as sess: sess.run(tf.global_variables_initializer())

constant_graph = \
    graph_util.convert_variables_to_constants(sess, sess.graph_def, ['op_to_store'])

print(sess.run(output, feed_dict={x: [2], y: [3]}))

with tf.gfile.FastGFile('./model.pb', 'wb') as f:
    f.write(constant_graph.SerializeToString())

############################################

pred = remote.execute

[sleepsonthefloor]

You can do it with remote.execute_multi:

from graphpipe import remote
import numpy as np
x = np.array([0, 1]).astype(np.int32)
y = np.array([0, 1]).astype(np.int32)
result = remote.execute_multi("http://127.0.0.1:9000", [x, y], None, None) # inputNames and outputNames are not specified, so defaults are inferred from the model structure

print(result)

Or, if you want to specify the names of the inputs and outputs associated with your call:

from graphpipe import remote
import numpy as np
x = np.array([0, 1]).astype(np.int32)
y = np.array([0, 1]).astype(np.int32)
result = remote.execute_multi(
        "http://127.0.0.1:9000",
        [x, y],                # inputs
        ['x', 'y'],             # inputNames
        ['op_to_store']) # outputNames
print(result)

That said, I don't see why remote.execute should not be able to handle this through type introspection. Perhaps we should allow remote.execute to detect if an array is passed in, and if so apply multiple inputs.

btw, I ran your model like this:

docker run -it --rm \
        -v "$PWD:/models/"  \
        -p 9000:9000 \
        sleepsonthefloor/graphpipe-tf:cpu \
        --model=/models/model.pb \
        --listen=0.0.0.0:9000

Generated the model like this:

import tensorflow as tf
x = tf.placeholder(tf.int32, shape=[1], name='x')
y = tf.placeholder(tf.int32, shape=[1], name='y')
b = tf.Variable(1, name='b')
xy = tf.multiply(x, y, name='multiply')
output = tf.add(xy, b, name='op_to_store')

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    constant_graph = \
        tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, ['op_to_store'])

    print(sess.run(output, feed_dict={x: [2], y: [3]}))

    with tf.gfile.FastGFile('./model.pb', 'wb') as f:
        f.write(constant_graph.SerializeToString())

[sleepsonthefloor]

https://github.com/oracle/graphpipe-py/pull/5 < perhaps this is worth considering

[wangce888]

@sleepsonthefloor 赞