/name-entity-recognition

Repo for CSE_517 assignment 4: Structured Learning with Feature-Rich Models

Primary LanguagePython

README.md file for CSE_517 assignment 4: Structured Learning with Feature-Rich Models.

###Run Best Model and print result

  1. Run source best_model.sh
  2. Run source eval.sh
  3. Run source read.sh

###Data Location

  1. First run cp config.ini.template config.ini
  2. Change the data_dir attributes in config.ini file to the data location you place.
  3. To simplified the procedure above, I have put the data dir inside this folder.

###Usage

  1. Use python main.py --help to see argument list you can use.
  2. Use --lrn-rate [LEARN_RATE] to specify learning rate eta of Perceptron.
  3. Use --epoch [EPOCH_TIME] to specify how many epoch the algorithms will run through.
  4. --no-ctoken, --no-ptoken, --no-ftoken, --no-cpos, --no-ppos, --no-fpos, --no-cchunk, --no-pchunk and --no-fchunk means no current token, no previous token, no future token, no current pos tag, no previous pos tag, no future pos tag, no current syntactic chunk, no previous syntactic chunk and no future syntactic chunk as features, respectively.
  5. Use --test to train the model with "train+dev" set and evaluate the model with "test" set. If not specified, it will be trained on "train" and evaluate the model with "dev" set.
  6. When --small is specified, the learning algorithms use the truncated (small) dataset.

###Output

  1. The output files will be located in the output folder with the filename:

    [dev|train]_eta[LEARN_RATE]_epoch_[EPOCH_TIME]_[FEATURE_CODE].txt

  2. Use ./conlleval.txt < filename to eval the result.

  3. FEATURE_CODE is a sequence of digits of length six. It's the true-false value of --no-ctoken, --no-ptoken, --no-ftoken, --no-cpos, --no-ppos, --no-fpos, --no-cchunk, --no-pchunk and --no-fchunk. Where 1 means True and 0 means False.