#Structured Reading

Implementation of structured reading active learning

How to run?

From the command line:

$ python experiment.py --train data_name --config config_file.cfg

data_name can be: imdb, sraa, 20news

Configuration File

Configuration file has the same format as an INI file. Sections [sections] and key=value or key:value. However, for this project values should reflec python syntax. For example, strings should be quoted, boolean should be capitalized.

The sections are:

• learner
• expert
• experiment
• data

Example:

[learner]
type           : "sequential"
model          : "lrl2"
parameter      : 1
utility      : "rnd"
snippet      : "sr"
                
			
[expert]        
type           : "sent"
model          : "lrl2"
parameter      : 1
costmodel      : 1
                

[data]
limit			: 2
split			: 0.5
min_size		: 10
vectorizer 		: "tfidf"
categories		: "religion"

[experiment]   
bootstrap      : 50
stepsize       : 10
maxiter        : 125
budget         : 1250
cheatingmode   : False
trials         : 5
folds          : 1
fileprefix     : "RND-F1-LIM2-"
outputdir      : "../results"
seed           : 876543210
limit          : 2
costfunction   : "unit"

The possible values for each section are:

• learner section

  • type: sequential, joint
  • model: lr, lrl2, mnb
  • utility: rnd, unc
  • snippet: sr, rnd, first1

• expert section

  • type: sent(ence), true(labels), pred(icting)
  • model: lr, lrl2, mnb

• data section

  • vectorizer: tfidf, bow
  • categories: only applies for some data. options: None, religion (20news)