Hi Francisco, can you please provide some info on how do you generate sparse vectors for words? How did you choose the length of the vectors (16k)?
Thanks! Michael On Mon, Feb 9, 2015 at 8:46 AM, cogmission1 . <[email protected]> wrote: > Hi Francisco, > > Those are really exciting developments! It's also very exciting to hear > that you've formalized your usage of HTM.java. I've just created a jmh > [1][2][3][4] Benchmark Tooling Apparatus for HTM.java so that we can now > look at formally starting to optimize the codebase and have legitimate > numbers from which to proceed! I wanted to get started on development of a > Network API to pull together the assembly of components, but I wanted to be > mindful of its impact on performance - which we can now do because of the > baseline Benchmarks! (Waiting to get merged now.) > > Exciting times! > > Regards > > David > > 1. http://openjdk.java.net/projects/code-tools/jmh/ > 2. Vital video presentation by the author: http://vimeo.com/78900556 > 3. More information: > http://www.oracle.com/technetwork/articles/java/architect-benchmarking-2266277.html > 4. Github Pull Request for HTM.java Benchmark Harness here: > https://github.com/numenta/htm.java/pull/180 > > On Mon, Feb 9, 2015 at 5:32 AM, Francisco Webber <[email protected]> wrote: > >> Hi All, >> Thank you for the promotion help Mika ;-) >> One of the reasons it took a while to publish the python client is that >> we use Swagger to generate it and had to create clean template code. >> Currently there are the Java, Python and PHP SDKs, more to follow soon. >> Another upcoming item is the first batch of our cross-lingual Retinas. We >> will soon open a beta-program for people who want to experiment with >> cross-lingual word-SDRs. The available languages are: English, French, >> Spanish, German, Danish, Chinese, Arabic, Russian with more to follow. >> The aim of the cross-lingual Retinas is to have homologue representations >> for the same concept in all the offered languages. >> The SDR for English “horse”, French “cheval” or German “Pferd” is always >> (nearly) the same. If you crate a Fingerprint of a text in one language, it >> can be directly compared with the Fingerprint of a text in another >> language. It should be possible to create interesting setups where an HTM >> network could learn to compare sequences of words in different languages… >> Another project that will be published in the coming weeks, is a >> collection of plugin-nodes for the Eclipse based Knime workflow system ( >> www.knime.org). The Knime-nodes will implement the cortical.io API >> functions and the current Java-HTM version of Nupic. By using Knime it >> should be much easier for non-programers to setup and exchange experiments. >> We will offer an automatic update site for the cortical and Nupic nodes. >> As for the white paper, it has been delayed due to a lot of biz-dev work >> in the past two months, but there will finally be a first version coming >> out by the end of February. >> >> Any feedback or questions from the Nupic community will of course be more >> than welcome. >> >> All the Best >> >> Francisco >> >> On 06.02.2015, at 16:06, Mika Schiller <[email protected]> wrote: >> >> Hello Nupic, >> >> In case you don't already know cortical's Python client is finally >> available. >> >> https://github.com/cortical-io/python-client-sdk >> >> Fransisco, thanks for you and your team's hard work on all this and do >> let us know when the draft version of the Semantic Folding Theory white >> paper is available. It would also be great to know about any significant >> NLP-related progress you guys and Numenta make moving forward. Perhaps you >> could update us from time to time. >> >> Mika >> >> >> > > > -- > *We find it hard to hear what another is saying because of how loudly "who > one is", speaks...* >
