The power_method in summarizers/lex_rank.py was not correct. The lambda_val was increasing for this test and hence resulted in inf and nan values reported in #187 (also a warning would be raised when running that test).
The problem was that the next_p vector should have been normalized after the matrix multiplication according to the Wikipedia entry, which it was not.
AryazE
commented
11 months ago
The
power_methodinsummarizers/lex_rank.pywas not correct. Thelambda_valwas increasing for this test and hence resulted ininfandnanvalues reported in #187 (also a warning would be raised when running that test). The problem was that thenext_pvector should have been normalized after the matrix multiplication according to the Wikipedia entry, which it was not.