You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Don't forget about `--copt` from above if boringssl fails to build.
40
44
41
45
To get options for training, use:
42
-
> bazel run //src/training/train --help
46
+
> bazel run //n2p/training:train --help
43
47
44
48
By default, train gets input programs (converted to JSON for example with UnuglifyJS) from the file testdata in the current directory. As a result, it creates files with the trained model.
45
49
46
50
If you wish to train the model using pseudolikelihood use the following parameters:
47
51
48
-
> bazel run //src/training/train -- -training_method pl -input path/to/input/file --logtostderr
52
+
> bazel run //n2p/training:train -- -training_method pl -input path/to/input/file --logtostderr
49
53
50
54
you can control the pseudolikelihood specific beam size with the `-beam_size` parameter which is different from the beam size used during MAP Inference.
51
55
52
-
`//src/training/train` expects data to be in protobuf recordIO format. If you want to use JSON input - use `//src/training/train_json` instead.
56
+
`//n2p/training:train` expects data to be in protobuf recordIO format. If you want to use JSON input - use `//n2p/training:train_json` instead.
53
57
54
58
### Factors
55
59
56
60
by default the usage of factor features in Nice2Predict is enabled, however if you wish to disable it you can launch the training with the following command:
57
61
58
-
> bazel run //src/training/train -- -use_factors=false -input path/to/input/file --logtostderr
62
+
> bazel run //n2p/training:train -- -use_factors=false -input path/to/input/file --logtostderr
0 commit comments