Text to SQL Queries

WikiSQL is one of the most popular benchmarks in semantic parsing. It is a supervised text-to-SQL dataset, beautifully hand-annotated by Amazon Mechanical Turk. Some of the early works on WikiSQL modeled this as a sequence generation problem using seq2seq but we are moving away from it. The text has to be cleaned before passing it to the model like doing decontraction of the words, removing stop words, removing non-alphanumeric text from the corpus. As we have the dataset in SQL queries and headers, so we have to featurize the text using a tokenizer from the nltk library and then concatenate the query and headers.