Using Tensorflow with pythonnet · Issue #473 · pythonnet/pythonnet · GitHub

Environment

Pythonnet version: 2.3.0
Python version:3.5.2 (Anaconda)
Operating System:Windows 10

Details

                                dynamic tf = Py.Import("tensorflow");
                               dynamic data = np.ones(5);
                                dynamic x = tf.constant(data, Py.kw("name", 'x'));

THIS LINE>> dynamic y = tf.Variable((x * x) * 5 - 3 * x + 15, Py.kw("name", 'y'));

An unhandled exception of type 'Microsoft.CSharp.RuntimeBinder.RuntimeBinderException' occurred in System.Core.dll

Additional information: Operator '*' cannot be applied to operands of type 'int' and 'Python.Runtime.PyObject'

@mphielipp i think this is the same as this issue, which is limitation of .NET DLR:

#86

You need to cast dynamic to corresponding type (e.g. double) when dynamic is NOT 1st argument:

#249 (comment)

I try different castings (dynamic, int, double) but does not work:

dynamic y = tf.Variable( (cast)3 * x ), Py.kw("name", 'y'));

An unhandled exception of type 'Microsoft.CSharp.RuntimeBinder.RuntimeBinderException' occurred in PythonNet2.exe

Additional information: Operator '*' cannot be applied to operands of type 'int' and 'Python.Runtime.PyObject'

Here is how casting works:

dynamic y = tf.Variable( 3 * ((int)x) ), Py.kw("name", 'y'));

I put extra brackets to disambiguate the operation ordering.

I typed:
dynamic y = tf.Variable((3 * ((int)x)), Py.kw("name", 'y'));
also tried
dynamic y = tf.Variable(3 * ((int)x), Py.kw("name", 'y'));

same error: (An unhandled exception of type 'Microsoft.CSharp.RuntimeBinder.RuntimeBinderException' occurred in ...Additional information: Cannot convert type 'Python.Runtime.PyObject' to 'int')

wait x is not an int, x was declare in the previous line:
dynamic x = tf.constant(data, Py.kw("name", 'x'));

Now I see the issue: trying to pass dynamic pythonnet expression to tensorflow. Let me think about this.

…

On Thu, May 25, 2017, 5:34 PM mphielipp ***@***.***> wrote: wait x is not an int, x was declare in the previous line: dynamic x = tf.constant(data, Py.kw("name", 'x')); — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#473 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHgZ5dR4DV-W-CepWqBAwtBWQhO7xea1ks5r9gHYgaJpZM4Nm72a> .

I don't have much experience with tensorflow. Can you point to any tutorial that does the same in Python API?

Can't you build your expressions only using tensorflow math functions, not Python operators?

https://www.tensorflow.org/api_guides/python/math_ops

Yes. This is a sample code:
import tensorflow as tf
import numpy as np

data = np.random.randint(1000, size=10000)

print (data)

x = tf.constant(data, name='x')
y = tf.Variable((x * x) * 5-3*x+15, name='y')

model = tf.global_variables_initializer()
with tf.Session() as session:
session.run(model)
print(session.run(y))

it seems that the operators are pass as Python ones in this example.

looks like possible using only tensorflow math:

activate py35

(py35) C:\Users\denis.akhiyarov>conda update python
(py35) C:\Users\denis.akhiyarov>pip install tensorflow
(py35) C:\Users\denis.akhiyarov>python
Python 3.5.3 |Continuum Analytics, Inc.| (default, May 15 2017, 10:43:23) [MSC v
.1900 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow
>>> import tensorflow as tf
>>> import numpy as np
>>> data = np.ones(5)
>>> x=tf.constant(data, name='x')
>>> y = tf.Variable((x * x) * 5 - 3 * x + 15, name='y')
>>> z = tf.Variable(tf.multiply(tf.pow(x,2),5), name='z')
>>> exit()

(py35) C:\Users\denis.akhiyarov>pip install pythonnet -U
(py35) C:\Users\denis.akhiyarov>set PYTHONHOME=C:\Python\Python27_64b\envs\py35
(py35) C:\Users\denis.akhiyarov>scriptcs
scriptcs (ctrl-c to exit or :help for help)

> #r C:\Python\Python27_64b\envs\py35\lib\site-packages\Python.Runtime.dll
> using Python.Runtime;
> dynamic sys;
> using (Py.GIL()) {sys=Py.Import("sys");}
> dynamic np;
> using (Py.GIL()) {np=Py.Import("numpy");}
> dynamic tf;
> using (Py.GIL()) {tf=Py.Import("tensorflow");}
> dynamic data = np.ones(5);
> dynamic x=tf.constant(data, Py.kw("name","x"));
> dynamic z = tf.Variable(tf.multiply(tf.pow(x,2),5),  Py.kw("name","z"));

ok. thanks!!!

            dynamic data = np.ones(5.0);
            dynamic x = tf.constant(data, Py.kw("name", 'x'));
            dynamic y = tf.Variable(tf.multiply(tf.constant(3) ,x), Py.kw("name", 'y'));

TypeError : Input 'y' of 'Mul' Op has type float64 that does not match type int32 of argument 'x'."}
x is type ndarray and each element if float 64

well i guess you need to specify the types:

dynamic y = tf.Variable(tf.multiply(tf.constant(3.0,Py.kw("dtype","float64")),x), Py.kw("name", 'y'));

Yes!! Thanks!!

I tried the specification but did not change the 3 to 3.0. Thanks again!

I did run:
dynamic x = tf.constant(data,Py.kw("dtype","float64"), Py.kw("name", 'x'));
dynamic y = tf.Variable(tf.multiply(tf.constant(3.0, Py.kw("dtype", "float64")), x), Py.kw("name", 'y'));
dynamic model = tf.global_variables_initializer();
dynamic res = tf.Session().run(model);
Console.WriteLine(tf.Session().run(y));

But it does not initialize properly:

{"FailedPreconditionError : Attempting to use uninitialized value y\n\t [[Node: _send_y_0 = _SendT=DT_DOUBLE, client_terminated=true, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/cpu:0", send_device_incarnation=3877403296068384333, tensor_name="y:0", _device="/job:localhost/replica:0/task:0/cpu:0"]]"}

dynamic ssn=tf.Session();
ssn.run(model);
ssn.run(y);

Great!! Thanks!!

            // w = tf.Variable([1.0, 2.0], name = "w")
            dynamic wa = np.array(new List<double> { 1.0, 2.0});
            dynamic w = tf.Variable(wa, Py.kw("dtype", "float64"), Py.kw("name", 'w'));

{"TypeError : float() argument must be a string or a number, not '0, Culture=neutral, PublicKeyToken=b77a5c561934e089]]'"}

But can you do the same in Python? Try skipping the dtype argument or also providing the shape argument. In this case TF may be able to resolve dtype and shape based on value.

Ok. Thanks.

On Fri, May 26, 2017 at 11:13 PM denfromufa ***@***.***> wrote: But can you do the same in Python? Try skipping the dtype argument or also providing the shape argument. In this case TF may be able to resolve dtype and shape based on value. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#473 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AECtkPbHhN9lGqy2F1GQ_H0CpwSen7AEks5r90BkgaJpZM4Nm72a> .

-- Best Regards, Mariano

@mphielipp here is the full code sample, I hope you don't mind that I used it in one presentation:

// RUN THIS IN SCRIPTCS
#r C:\Python\Python35_64b\lib\site-packages\Python.Runtime.dll
using Python.Runtime;
dynamic sys; using (Py.GIL()) {sys=Py.Import("sys");}
dynamic np; using (Py.GIL()) {np=Py.Import("numpy");}
dynamic tf; using (Py.GIL()) {tf=Py.Import("tensorflow");}
dynamic data = np.ones(5);
dynamic x=tf.constant(data, Py.kw("name","x"));
dynamic z = tf.Variable(tf.multiply(tf.pow(x,2),5),  Py.kw("name","z"));
dynamic y = tf.Variable(tf.multiply(tf.constant(3.0,Py.kw("dtype","float64")),x), Py.kw("name", 'y'));
dynamic model = tf.global_variables_initializer();
dynamic ssn=tf.Session(); ssn.run(model); ssn.run(y).ToString();

@denfromufa
Because you were very active in the past, I adress you directly. Maybe you can help me. Thanks a lot in advance.

I have installed "python3.6" over anaconda3 and added "tensorflow" with the published cmd operations.
It works fine and it is possible to run the application (for instance, var tf=Py.Import("tensorflow"); ...)
But I get a failure b< one of you last statements in the above example.

2018-09-27 13:36:18.984929: I T:\src\github\tensorflow\tensorflow\core\platform\cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2

Do you have any idea?

That is a warning, not an error

ok! thank's for the answer.

@denfromufa
I am searching in the internet for an example, that uses keras over pythonnet. You wrote in an other thread a year ago, that it is possible? I have no luck by finding a syntax in c# for this. Do you have any sample code? Best regards, Daniel

Using Tensorflow with pythonnet · Issue #473 · pythonnet/pythonnet · GitHub

Repository navigation

Using Tensorflow with pythonnet #473

Environment

Details

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Using Tensorflow with pythonnet · Issue #473 · pythonnet/pythonnet · GitHub

Repository navigation

Using Tensorflow with pythonnet #473

Description

Environment

Details

Activity

den-run-ai commented on May 25, 2017

mphielipp commented on May 25, 2017

den-run-ai commented on May 25, 2017

mphielipp commented on May 25, 2017

mphielipp commented on May 25, 2017

den-run-ai commented on May 25, 2017

den-run-ai commented on May 26, 2017

mphielipp commented on May 26, 2017

mphielipp commented on May 26, 2017

den-run-ai commented on May 26, 2017

mphielipp commented on May 26, 2017

mphielipp commented on May 26, 2017

den-run-ai commented on May 26, 2017

mphielipp commented on May 26, 2017

mphielipp commented on May 26, 2017

mphielipp commented on May 26, 2017

den-run-ai commented on May 26, 2017

mphielipp commented on May 26, 2017

mphielipp commented on May 26, 2017

den-run-ai commented on May 26, 2017

mphielipp commented on May 28, 2017

den-run-ai commented on Sep 14, 2017

danieljaro commented on Sep 27, 2018

den-run-ai commented on Sep 27, 2018

danieljaro commented on Sep 27, 2018

danieljaro commented on Sep 28, 2018

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions