Hi guys, First of all, I have to mention, that I have almost no experience with kind of models. From tensorflow I have been looking a bit at tensorflow probability[1]. Could imagine that also v.class.ml[2] and r.learn.ml[3] could give some inspiration how such libraries could be plugged into GRASS in principle. Other interesting modelling libraries would be PyMC3[4] or Stan[5]. Just for mentioning it.
I can imagine that IO is particularly important for parallelism in these libraries... Maybe it would be possible develop a general framework for this? Cheers Stefan 1: https://www.tensorflow.org/probability 2: https://grass.osgeo.org/grass74/manuals/addons/v.class.ml.html 3: https://grass.osgeo.org/grass74/manuals/addons/r.learn.ml.html 4: https://docs.pymc.io/ 5: https://mc-stan.org/ -----Original Message----- From: grass-dev <grass-dev-boun...@lists.osgeo.org> On Behalf Of Maris Nartiss Sent: tirsdag 5. mars 2019 18:00 To: Luca Delucchi <lucadel...@gmail.com> Cc: Soumya kanta Das <dassoumyakant...@gmail.com>; GRASS-dev <grass-dev@lists.osgeo.org> Subject: Re: [GRASS-dev] Introduction and Idea ( GSOC 2019 ) Hello, svētd., 2019. g. 3. marts, plkst. 08:50 — lietotājs Luca Delucchi (<lucadel...@gmail.com>) rakstīja: > > On Sat, 2 Mar 2019 at 18:48, Soumya kanta Das > <dassoumyakant...@gmail.com> wrote: > > > > Hello, > > Hi, > > > Myself Soumya Kanta Das, studying Geo-informatics. Currently i work on > > automating geospatial workflow and Machine learning applications. I have > > previously worked on automated identification of buildings in Satellite > > imagery using Deep learning. I have language proficiency in python. > > > > I would like to develop a Deep learning module for grass using tensorflow > > library and python 3. Which would be very helpful for end-users. > > > > something already exists in GRASS GIS addons using tensorflow [0], you > can have look there.. I have a working prototype of GRASS – Keras bridge (still needs a lot of work to fix all edge cases + support of second image_data_format). My first approach was to use pure Python code, but it turned out to be realllly slooow. Large parts have to be written in C (linked via ctypes). Currently there still is a speed penality due to GRASS raster code not supporting mutlithreaded applications. It all can be worked around. Developing any high speed interface between ML libraries and GRASS needs some (good) understanding of working with multidimensional arrays in C and GRASS rasters in C. > > Thank you. > > > > [0] https://grass.osgeo.org/grass74/manuals/addons/i.ann.maskrcnn.html > > -- > ciao > Luca Just FYI, Māris. _______________________________________________ grass-dev mailing list grass-dev@lists.osgeo.org https://lists.osgeo.org/mailman/listinfo/grass-dev _______________________________________________ grass-dev mailing list grass-dev@lists.osgeo.org https://lists.osgeo.org/mailman/listinfo/grass-dev
