About Prior class distribution parameter (%true PIs)?
Prior class distribution parameter represents an estimate of the distribution of true PIs in the population from where the target PI text samples are drawn. Ideally, this parameter should be the same for training/test data but that may rarely happen in practice. Since this value can only be roughly estimated, one can start with the baseline estimates calculated from their training/test datasets. Subsequently, this parameter may be tuned/explored further with other different values (by comparing with baseline estimates) to get a good system performance on his dataset (e.g. in terms of F-measure). Those users who wish to do large scale analysis on PubMed abstracts, this value may be estimated from annotated samples that are generated randomly from the PubMed abstracts. A crude estimate of this value for analyzing randomly generated (or a large scale) samples from PubMed abstracts may be somewhere in the region of 10% (a rough guesstimate). If this parameter is unchecked (default setting)
Related Questions
- If I purchased Class C Shares which had an 18 month hold prior to the change, is my client still required to hold this position for 18 months or does the holding period drop to 12 months?
- May I release for distribution in-date components that were collected prior to instituting enhanced inspection but were leukoreduced and have not been transfused?
- What is the requirement for a statistics class prior to NUR 600 class?
