Hi Bruce. Thanks for the feedback! I just pushed code changes that fix the first issues you had. The
get_extracted_sequence() bug was a bad merge on my part. The unknown keyword is a Keras version issue: the code was written for Keras 1.X, and you have Keras 2.X installed. I’ve updated the code everywhere to use Keras 2 APIs and added a
As for the OOM issue: What settings are you running in
train.py? Specifically this part:
model = 'lstm' # see `models.py` for more
saved_model = None # None or weights file
class_limit = None # int, can be 1-101 or None
seq_length = 40
load_to_memory = True # pre-load the sequences into memory
In the latest push, I’ve reduced the number of units in the LSTM which should help. I trained this network on a 12 gb GPU. Not sure how much memory the 1080s have.
Thanks again for pointing out the issues.