Tensorflow session in a class in python?

Question

I want to use tensorflow gradients for a computation of other quantities later on. I need to numerically compute the objective function and gradients as functions in a class (This class then is used in the remaining suite). However, I am getting error for the below code:

import tensorflow as tf
class MyClass:
    def __init__(self):
        x=tf.Variable(tf.zeros(2))
        func = tf.cos(14.5 * x[0] - 0.3) + (x[1] + 0.2) * x[1] + (x[0] + 0.2) * x[0]
        diff_func = tf.gradients(func,x)

        sess = tf.Session()

    def getFunc(self,coords):
        return self.sess.run(self.func,feed_dict={self.x:coords})

    def getGrad(self,coords):
        grad = self.sess.run(self.diff_func,feed_dict={self.x:coords})
        return grad

MyClass = MyClass()
MyClass.getFunc([0.362,0.556])
MyClass.getGrad([0.362,0.556])

The error I am getting is:

Traceback (most recent call last):

File "", line 19, in MyClass.getFunc([0.362,0.556])

File "", line 11, in getFunc return self.sess.run(self.func,feed_dict={self.x:coords})

AttributeError: MyClass instance has no attribute 'sess'

Not sure how I can get this class running correctly. Thanks.

You may define `sess = tf.Session()` in top outside this class. — Dinusha Thilakarathna, Mar 30 '18 at 05:44

score 5 · Answer 1 · answered Mar 30 '18 at 00:47

Basically, 'self' tells which variables and methods belongs to a class. So you have to tell that (x, func, diff_func and sess) belong to the MyClass. So modify the code as below:

import tensorflow as tf


class MyClass:
    def __init__(self):
        self.x = tf.Variable(tf.zeros(2))
        self.func = tf.cos(14.5 * self.x[0] - 0.3) + (self.x[1] + 0.2) * self.x[1] + (self.x[0] + 0.2) * self.x[0]
        self.diff_func = tf.gradients(self.func, self.x)

        self.sess = tf.Session()

    def getFunc(self, coords):
        return self.sess.run(self.func, feed_dict={self.x: coords})

    def getGrad(self, coords):
        grad = self.sess.run(self.diff_func, feed_dict={self.x: coords})
        return grad


MyClass = MyClass()
MyClass.getFunc([0.362, 0.556])
print(MyClass.getGrad([0.362, 0.556]))

score 1 · Answer 2 · answered Mar 29 '18 at 22:30

1

Replace sess = tf.Session() with self.sess = tf.Session().

answered Mar 29 '18 at 22:30

Christian NH

71
2

Didn't work. Still the error "Traceback (most recent call last): File "", line 19, in MyClass.getFunc([0.362,0.556]) File "", line 11, in getFunc return self.sess.run(self.func,feed_dict={self.x:coords}) AttributeError: MyClass instance has no attribute 'func'" – Dave Mar 29 '18 at 22:34
1

This problem is solved a similar way. You must write `self.func` if you want `func` to be accessible anywhere in the class. In the snippet you've posted, the variables defined in `__init__` are only accessible from within `__init__`. [This post](https://stackoverflow.com/a/625097/4030272) elaborates on how it works. On a broader note, learning Python and TensorFlow by themselves is a challenge. Learning them at the same time is a recipe for frustration. You can save yourself some hassle by getting the fundamentals of Python down before moving on to TensorFlow. Happy hacking! – Christian NH Mar 29 '18 at 22:59
I have a very specific question. I have carved out a minimal code where the problem can be reproduced. Wouldn't it be more efficient to provide the corrected full working code? Then, I can build on it for more complicated cases. – Dave Mar 29 '18 at 23:06

score 0 · Answer 3 · answered Jul 21 '19 at 14:05

If you are creating a variable inside constructor viz __init__. Then you need to associate self for showing that this variable belongs to that particular class. In this case, __init__ is the constructor of MyClass, then for defining sess, you need to associate self before it such as self.var_name. So that class functions getFunc() and getGrad() can understand that sess variable is an object variable and can use it. So just replace self with self.sess in the constructor.

Tensorflow session in a class in python?

3 Answers3