Hi Everybody, I am trying to find a model for a dataset using the swarming algorithm. My problem is the following: There is a sequence of input parameters that are supposed to be mapped to a label. How can I express this in the search_def.json file? My current swarming results show no field contribution for all fields and disabled spatial and temporal pooler, although I believe this problem to be spatiotemporal in nature.
To illustrate this some more: [...] 2014-01-01 00:00:01,h#,0.591832,-0.694726,-0.803458,-0.895434,-0.848500 2014-01-01 00:00:02,h#,0.350835,-0.754058,-0.794732,-0.889368,-0.854815 2014-01-01 00:00:03,h#,0.348494,-0.709733,-0.796168,-0.894366,-0.849283 [...] 2014-01-01 00:00:38,hv,-0.269301,-0.625990,-0.725829,-0.238063,1.387178 2014-01-01 00:00:39,hv,-0.368703,-0.630025,-0.761165,-0.344434,1.662843 2014-01-01 00:00:40,hv,0.116414,-0.601504,-0.732410,-0.409445,1.894189 2014-01-01 00:00:41,hv,-0.285989,-0.595255,-0.762528,-0.583831,1.813830 [...] This is a possible input. The length of each sequence is unknown, but the label demarcates the start and end points. The first three lines belong to sequence 'h#' while the bottom four lines belong to sequence 'hv'. I was also wondering whether it is possible to have a "supervised mapper" as in [1] or something similar to train the model, i.e. insert sequence as input to the model and then insert label? [1] J. Doremalen - Spoken Digit Recognition using a Hierarchical Temporal Memory All the best, Manuel _______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
