Dfyjkt

I have a tfrecord file which should have ~8000 image patches in it, all 128x128x4. I've written an input function about the same as what I found on the tensorflow site here:
https://www.tensorflow.org/guide/datasets#randomly_shuffling_input_data

My code to read the tfrecord is:

def tfrecord_input_fn(fileName,
 numEpochs=None,
 shuffle=None,
 batch_size=None):
dataset = tf.data.TFRecordDataset(fileName, compression_type='GZIP')
dataset = dataset.map(lambda example_proto: tf.parse_single_example(example_proto, featuresDict))
if shuffle == True:
 dataset = dataset.shuffle(buffer_size=(batch_size * 10))
dataset = dataset.batch(batch_size)
dataset = dataset.repeat(numEpochs)
iterator = dataset.make_one_shot_iterator()
features = iterator.get_next()
return features

When I train it on the network that I've written everything seems to work... however I've noticed that when I'm running it in a session, it only appears to use a very small subset of the input images. I'm using it like this:

with tf.Session() as sess:
 for ep in range(num_epochs):
 train_img_lab = tfrecord_input_fn(fileName=datapath, numEpochs=epoch_num, shuffle=True, train_batch_size=train_batch_size)

 a = sess.run(train_img_lab)
 image, label = create_np_arrray_from_tensor(a, train_batch_size)
 _, c = sess.run([train_op, train_loss], feed_dict= trainimg_placeholder: image, trainlabels_placeholder: label})

When I wasn't seeing the outputs from the model that I expected I wrote something to save the 128X128 images to my disk. That's when I realized that somehow I was using the same 10 or 15 images of the ~8000 total each batch of each epoch. I suspect this means I'm either not understanding something in my tfrecord_input function (maybe the batch or repeat functions?) or I'm not calling it at the correct point in the session. Could someone help me understand the steps I'm taking, and the steps I need to take, to actually use all the images in my training data?

Thank you!