Cannot interpret feed_dict key as Tensor for batch and test

Question

I'm a newbie in Tensorflow and trying to do MNIST example in python 3.6.

I built up the code to see how errors changes during # of iterations, but it gives me an error on feed_dict inside of sess.run.

Here is my code as below,

import tensorflow as tf
import numpy as np
import functools
import sys
sys.path.append('./utils')

from mnist import MNIST

def lazy_property(function):
    attribute = '_cache_' + function.__name__

@property
@functools.wraps(function)
def decorator(self):
    if not hasattr(self, attribute):
        setattr(self, attribute, function(self))
        return getattr(self, attribute)

    return decorator

class Model:

  def __init__(self, image, label):
    self.image = image
    self.label = label
    self.logits
    self.prediction
    self.optimize
    self.error


  @lazy_property
  def logits(self):
    weight = tf.Variable(tf.zeros([img_size_flat, num_classes]))
    #print (img_size_flat)
    biases = tf.Variable(tf.zeros([num_classes]))
    #print (num_classes)
    equation = tf.matmul(self.image, weight) + biases
    return equation

  @lazy_property
  def prediction(self):
    return tf.nn.softmax(self.logits)

  @lazy_property
  def optimize(self):
    cross_entropy = tf.nn.softmax_cross_entropy_with_logits_v2(logits=self.logits, labels=self.label)
    cost = tf.reduce_mean(cross_entropy)
    return tf.train.GradientDescentOptimizer(learning_rate=0.5).minimize(cost)


  @lazy_property
  def error(self):
    y_pred_cls = tf.argmax(self.prediction, axis=1)
    mistakes = tf.not_equal(y_true_cls, y_pred_cls)
    #print(mistakes)
    error_from_acc = tf.reduce_mean(tf.cast(mistakes, tf.float32))
    return error_from_acc


batch_size = 100
num_steps = 1000

tf.reset_default_graph()

data = MNIST(data_dir="data/MNIST/")
img_size_flat = data.img_size_flat
img_shape = data.img_shape
num_classes = data.num_classes
image = tf.placeholder(tf.float32, [None, img_size_flat])
label = tf.placeholder(tf.float32, [None, num_classes])
y_true_cls = tf.placeholder(tf.int64, [None])
#print (y_true_cls)
model = Model(image, label)


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

    for step in range(num_steps):
        error = session.run(model.error, {x: data.x_test, y_true: data.y_test}) # Gives me an error message from HERE!!!!
        x_batch, y_true_batch, _ = data.random_batch(batch_size=batch_size)
        session.run(model.optimize, {x: x_batch, y_true: y_true_batch})
        if (step % 100 == 0):
            print("Error rate @ iter %d : %f" % (step, error))

What did I do wrong?

Where should I put that feed_dict ({x: data.x_test, y_true: data.y_test} and {x: x_batch, y_true: y_true_batch})?

Also, is there anywhere I did wrong in the code?

Please help me out here..

Thanks.

Answer 1

You have defined your placeholders like this:

image = tf.placeholder(tf.float32, [None, img_size_flat])
label = tf.placeholder(tf.float32, [None, num_classes])

, but then you are passing x and y_true as placeholders into session.run:

session.run(model.error, {x: data.x_test, y_true: data.y_test})

Therefore you need to replace x and y_true by image and label and you should be fine:

session.run(model.error, {image : data.x_test, label : data.y_test})