README

Here, BestTrees (see https://github.com/tm11ajn/besttrees/ for instructions of how to use the software) is tested.

The experiments are divided into directories according to which set of weighted tree automata (wta) was used. In each directory you will find a set of wta along with results of experiments run on that wta set. The results are in the form of plots, both in .png and .tex format. If you wish to have a look at the raw data, please see the .tex files.

A list of explanations:

* "wta_b" - the wta the generated wta set is based on.
* "1-f" - In this set, there are f files that were generated ...
* "k states" - ... based on a wta of k states.
* "All different weights" - all of the transition rules in the wtas have different weights.
* "Equal leaf weights" - all of the transition rules that are of rank > 0 have the same weight (namely 0) and the rank 0 rules have weight equal to the number of states.
* "Normal counter" - the transition rules are of the form a[q_k, q_j] -> q_i # (ijk / 100) and are added in the order of the weights (smallest weight first). (Note that ijk refers to concatenation, not multiplication.)
* "Backwards normal counter" - as normal counter, but the transition rules are added in reverse (largest weight first).
* "Mod [= modified] counter" - the counter has been modified to result in another order and to use an offset (please see the wta files for resulting format). 

The wta sets from the paper "A Comparison of Two N-Best Extraction Methods":

* Basic Example = example_wta_from_paper_unfolded
* Different Weights = wta_b_gen1-16_4_states_all_different_weights_tropical_semiring_backwards_normal_counter
* Modified Different Weights = wta_b_gen1-16_4_states_all_different_weights_tropical_semiring_mod_counter
* Equal Weights = wta_b_gen1-16_4_states_equal_leaf_weights_tropical_semiring_normal_counter