HazyResearch / deepdive

DeepDive
deepdive.stanford.edu
1.96k stars 536 forks source link

Does the latest deedive support incremental training (online learning) ? #534

Open RominYue opened 8 years ago

RominYue commented 8 years ago

I found an older version 0.6.0 mentioned incremental

but in the latest version, I could not found any documentation about incremental / online training.....but I thought the use of SGD in inference stage could support this.............

RominYue commented 8 years ago

Another problem when I just reuse the weight.

After pre-processing of NLP step, I type in the following command:

$ deepdive model weights reuse FOO

and the step goes forward to combine_factorgraph, but then

$ deepdive model infer

occurs the following error:

2016-05-07 12:50:49.932967 + mkdir -p probabilities
2016-05-07 12:50:49.935240 + mv -f weights/inference_result.out.text probabilities/
2016-05-07 12:50:49.938343 mv: cannot stat ‘weights/inference_result.out.text’: No such file or directory
‘run/ABORTED’ -> ‘20160507/125042.142224488’

The detail info shows as below: 1 2 3

alldefector commented 8 years ago

@RominYue What version of DD were you running (deepdive version)? Could you try the same on the latest version (git pull; git submodule update --init; make build; make install) and see if the error persists?