First of all, I'm confused as of how the save_dir works: if I train a network pointing to a directory, and then eventually train another one pointing to that same one, it seems like it overwrites properly the older data, but does not clear the extra-data from the older model. I think this is what caused the assert saved_chars == data_loader.chars, "Data and loaded model disagree on character set!" to trigger.
Secondly, is this a correct way of using the command: python sample.py -n=10000 --save_dir=testSave ?
Thirdly, what is the prime attribute of the sample ? And why does python sample.py take so long to execute?
Lastly, I'm wondering if I'm using the code properly since I have only downloaded the CPU version of TensorFlow. However, your documentation mentions that I we are supposed to add some lines of code, but not where: would you mind explaining a bit more the process if I want to use the CPU way? :)
Thank you so much! This is pretty neat. Currently teaching the Network to come up with fake laws, haha.
First of all, I'm confused as of how the
save_dir
works: if I train a network pointing to a directory, and then eventually train another one pointing to that same one, it seems like it overwrites properly the older data, but does not clear the extra-data from the older model. I think this is what caused theassert saved_chars == data_loader.chars, "Data and loaded model disagree on character set!"
to trigger.Secondly, is this a correct way of using the command:
python sample.py -n=10000 --save_dir=testSave
?Thirdly, what is the
prime
attribute of thesample
? And why doespython sample.py
take so long to execute?Lastly, I'm wondering if I'm using the code properly since I have only downloaded the CPU version of TensorFlow. However, your documentation mentions that I we are supposed to add some lines of code, but not where: would you mind explaining a bit more the process if I want to use the CPU way? :)
Thank you so much! This is pretty neat. Currently teaching the Network to come up with fake laws, haha.