Jumat, 05 Oktober 2012
Google ReFr Is a New Open Source Tool Designed to Tackle "Reranking" Problems
Google is not done open sourcing software designed to handle big data and tasks. This time around, it's not just getting rid of software it can't support anymore. Instead, it's releasing and open-sourcing a tool its Research team developed to handle reranking problems.
"Most speech and natural language processing models produce a set of hypotheses for each input, where each hypothesis has a score," Google Research explained.
"A reranker is any trained model that takes such a set of hypotheses and produces different scores that potentially produce a new ranked order, with the goal of improving the measurable quality of the system," it said.
Reranker Framework, or ReFr, is the name of the tool developed by a couple of Google scientists and which is now being made available.
The tasks ReFr is designed to counter target a very nice audience, but the tool should be really helpful for those dealing with this type of problems.
The core tool is built with C++, for performance, but it also comes with a runtime configuration mechanism and the ability to run the tasks in a Hadoop cloud, for the best performance.
Via: Google ReFr Is a New Open Source Tool Designed to Tackle "Reranking" Problems