search

TWOEARS / localisation-training-pipeline

Training of the localisation knowledge sources of the blackboard system
GNU General Public License v3.0
1 stars 1 forks source link

Setting of data directory for learned data #1

Closed hagenw closed 8 years ago

hagenw commented 8 years ago

The output directory is at the moment specified in get_data_root().

I would propose to remove this function completely and store the data directly under the corresponding data folder, which can be accessed via xml.dbPath. @ningma97 what do you think?

This is related to #2.

/CC @fietew @ryanckim @Hardcorehobel

ningma97 commented 8 years ago

Fine by me. The get_data_root() is really a dirty trick to run the code in our grid computing system without having to use the WP1 code.

hagenw commented 8 years ago

Ok, but your grid computing is a valid point. At the end we should ensure, that we are still possible to run the code on a cluster without too much hassle.

hagenw commented 8 years ago

The data is now stored under ./learned_models/DnnLocationKS and ./learned_models/GmmLocationKS. This behavior is identical to the one from the old and now removed GmtkLocationKS, see http://docs.twoears.eu/en/1.0/examples/train-localisation/. It has the advantage, that the Binaural Simulator first looks at a relative path for any data.

hagenw commented 8 years ago

The temporary feature folder (DevFeatures_1deg_32channels and TrainFeatures_MCT-DIFFUSE_1deg_32channels) are now stored in the temporary directory of the Binaural Simulator, that is defined by xml.dbTmp. The training pipeline depends now on twoears-main and the binaural simulator, but it should still be possible to use it on any kind of server, as the mex-files are not needed.