{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/character-level-recurrent-neural-networks-in","title":"Character-level Recurrent Neural Networks in Practice: Comparing Training and Sampling Schemes","arxiv_id":"1801.00632","date":"2018-01-02","proceeding":null,"authors":["Cedric De Boom","Thomas Demeester","Bart Dhoedt"],"abstract":"Recurrent neural networks are nowadays successfully used in an abundance of\napplications, going from text, speech and image processing to recommender\nsystems. Backpropagation through time is the algorithm that is commonly used to\ntrain these networks on specific tasks. Many deep learning frameworks have\ntheir own implementation of training and sampling procedures for recurrent\nneural networks, while there are in fact multiple other possibilities to choose\nfrom and other parameters to tune. In existing literature this is very often\noverlooked or ignored. In this paper we therefore give an overview of possible\ntraining and sampling schemes for character-level recurrent neural networks to\nsolve the task of predicting the next token in a given sequence. We test these\ndifferent schemes on a variety of datasets, neural network architectures and\nparameter settings, and formulate a number of take-home recommendations. The\nchoice of training and sampling scheme turns out to be subject to a number of\ntrade-offs, such as training stability, sampling time, model performance and\nimplementation effort, but is largely independent of the data. Perhaps the most\nsurprising result is that transferring hidden states for correctly initializing\nthe model on subsequences often leads to unstable training behavior depending\non the dataset.","url_abs":"http://arxiv.org/abs/1801.00632v2","url_pdf":"http://arxiv.org/pdf/1801.00632v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"character-level-recurrent-neural-networks-in","repo_url":"https://github.com/cedricdeboom/character-level-rnn-datasets","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null},{"paper_slug":"character-level-recurrent-neural-networks-in","repo_url":"https://github.com/tdmeeste/TimeAwareRNN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":"recommendation-systems","task_name":"Recommendation Systems"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}