nerudai | Czech poems generating system based on LSTM and Char-rnn

This respository includes code for training a neural net, generating poems and printing using thermal printer compatible with escpos drivers. After cloning repository use the included py36.txt to create conda repository with compatible packages. It is important to add this python-escpos package manually. After you put your data text file in data folder and change the path in train.py you first start the makeaplhabet.py to create aplhabet for Czech and Slovak Languages. If you need more characters, add them in any way to aplhabet file in data folder. The train script then uses only letters that occur in the alphabet file and ignores all the rest. This helped a lot during training as the raw scraped data was consisting of 450+ characters and after restricting this number to circa 150 the accuracy got much better. Training can be done by starting train.py and sampling by running sample.py. The client and server scripts were written in golang by my colleague Matus Bafrnec and I will ask him if it would be possible to release the source code as well.