CODEWITHSUNDEEP
pythonnumpytensorflow
Mewtwo
Mewtwo

Reputation: 1311

Execution stuck when evaluating TensorFlow tensors

So my code is:

with tf.Session() as sess:    
    init.run()
    epoch = 1
    iteration = 1
    print("Checkpoint 1")
    X_batch, y_batch = tf.train.batch([X_train, y_train], batch_size=batch_size)
    print("Checkpoint 2")
    X = X_batch.eval()
    y = X_batch.eval()
    print("Checkpoint 3")

The problem is that when executing the 2 last lines the execution stuck, not giving any output (Only checkpoint 1 & 2 are printed). I've searched across google and the conversion from tensorflow.python.framework.ops.Tensor to numpy.ndarray seems to be a trivial operation .

I also tried the following in case it makes some difference: 

X = X_batch.eval(session = sess)
y = X_batch.eval(session = sess)

Edit: I tried using Interactive session and the problem remains

`sess = tf.InteractiveSession()
 X_batch, y_batch = tf.train.batch([X_train, y_train], batch_size=batch_size)
type(X_batch)
type(y_batch)
print(type(X_batch.eval()))
sess.close()`

Upvotes: 2

Views: 1102

Answers (1)

pfm
pfm

Reputation: 6328

You need to start the queue runner hidden in tf.train.batch by using for example a tf.train.Coordinator (have look to this or this for example to get more insight). As per the docstring of the method (highlights from me):

This function is implemented using a queue. A QueueRunner for the queue is added to the current Graph's QUEUE_RUNNER collection.

and

The returned operation is a dequeue operation and will throw tf.errors.OutOfRangeError if the input queue is exhausted. If this operation is feeding another input queue, its queue runner will catch this exception, however, if this operation is used in your main thread you are responsible for catching this yourself.

As you didn't start the queue runner, the thread get stuck waiting for the enqueue operation to be run.

    X_batch, y_batch = tf.train.batch([X_train, y_train], batch_size=batch_size)
    with tf.Session() as sess:
        sess.run(init)
        coord = tf.train.Coordinator()
        threads = tf.train.start_queue_runners(coord=coord)
        try:
            while not coord.should_stop():
                X, y = sess.run([X_batch, y_batch])
        except Exception as e:
            coord.request_stop(e)
        finally:
            coord.request_stop()
            coord.join(threads)

Upvotes: 1

Related Questions