udacity / carnd-lenet-lab Goto Github PK

To solve this locally, I replaced the line:
saver.save(sess, 'lenet')
with
saver.save(sess, '.\lenet')

conda env create -f environment.yml fails because environment.yml is not in the repo

If you are going to pull files from the repo, please update the instructions in the README.md file.

ValueError: Parent directory of ../AI/tf/realuod_15_5_tf doesn't exist, can't save.

Traceback (most recent call last):
File "/home/atul/predictBO/tensorflow/makeAI-tf.py", line 151, in
learner.saveall()
File "/home/atul/predictBO/tensorflow/learning_util.py", line 167, in saveall
self.savemodel()
File "/home/atul/predictBO/tensorflow/learning_util.py", line 155, in savemodel
self.saver.save(self.sess, self.outname)
File "/home/atul/.local/lib/python3.6/site-packages/tensorflow/python/training/saver.py", line 1450, in save
raise exc
ValueError: Parent directory of ../AI/tf/realuod_15_5_tf doesn't exist, can't save.

Hello,

The code provided in the solution file will not run on a windows GPU enabled machine. Jupyter notebook crashes and gives non-useful error message. Running in normal python build yields the error: CUDA_ERROR_ILLEGAL_ADDRESS. This can be solved as suggested here: https://carnd-forums.udacity.com/questions/32717096/tensorflow-gpu-error-cudaerrorillegaladdress

For this instance, in[7] this works:

with tf.device('/cpu:0'):
	one_hot_y = tf.one_hot(y, 10)

As it can be a frustrating error, and jupyter gives no useful feedback, suggest offering this fix in the readme comments, or even updating the code until CUDA/tensorflow fix it properly.

Anthony

cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits, one_hot_y)
need to be:
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=one_hot_y)

If not we get:

Raise en error if labels do not sum to 1 exception

This came up in the discussion forum.