From d0a7c74b5d5e8aaf53d3241e09f1bda75bf0e1cc Mon Sep 17 00:00:00 2001 From: Chris Dyer Date: Mon, 11 Feb 2013 21:14:10 -0500 Subject: Baum Welch training for HMMs --- training/crf/baum_welch_example/README.md | 32 +++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+) create mode 100644 training/crf/baum_welch_example/README.md (limited to 'training/crf/baum_welch_example/README.md') diff --git a/training/crf/baum_welch_example/README.md b/training/crf/baum_welch_example/README.md new file mode 100644 index 00000000..97525da5 --- /dev/null +++ b/training/crf/baum_welch_example/README.md @@ -0,0 +1,32 @@ +Here's how to do Baum-Welch training with `cdec`. + +## Set the tags you want. + +First, set the number of tags you want in tagset.txt (these +can be any symbols, listed one after another, separated +by whitespace), e.g.: + + C1 C2 C3 C4 + +## Extract the parameter feature names + + ../mpi_extract_features -c cdec.ini -t train.txt + +If you have compiled with MPI, you can use `mpirun`: + + mpirun -np 8 ../mpi_extract_features -c cdec.ini -t train.txt + +## Randomly initialize the weights file + + sort -u features.* | ./random_init.pl > weights.init + +## Run training + + ../mpi_baum_welch -c cdec.ini -t train.txt -w weights.init -n 50 + +Again, if you have compiled with MPI, you can use `mpirun`: + + mpirun -np 8 ../mpi_baum_welch -c cdec.ini -t train.txt -w weights.init -n 50 + +The `-n` flag indicates how many iterations to run for. + -- cgit v1.2.3