22.9.2021 meeting

[teddy4445]

Teddy

• [ ] Investigate the connection between a "stable" and a "good" summary.
• [x] Implement a Greedy algorithm that takes a row + col in each iteration.
• [x] replacing the current "row" and "column" score with metric(m[t-1], n[t]) and metric(m[t], n[t-1]).
• [x] Test the Las-Vegas algorithms on the rows optimization only and compare with the Greedy algorithm (for the rows) again.
• [x] Try to divide the experiments into sub-folders with some logic (time, scores, summaries)
• [x] Add a "stability test" where the noise is defined by picking n<<N'<N, m<<M'<M in random.
• [x] In the score summary files, add "global score" columns.

Amit

• [ ] Implement (in Python) a "coverage" metric for a summary.
• [ ] Checkout with the Ph.D. student regarding the multi-arm bandit implementation and when she finished, share the code

[teddy4445]

The results of the task "Test the Las-Vegas algorithms on the rows optimization only and compare with the Greedy algorithm (for the rows) again":

algorithm,dataset_2_headech_prodrom,dataset_3_liver-disorders,dataset_4_mfeat-morphological,dataset_5_page-blocks
las_vegas,0.6784669059521363 \pm 0.6393907670465584,13.69597430449523 \pm 3.965459153090312,3.4150496618576685 \pm 3.2386512113354082,3.6077477390851147 \pm 1.8877882641390702
greedy,0.0 \pm 0.0,0.0 \pm 0.0,0.0 \pm 0.0,0.0 \pm 0.0

These results for 200X5 -> 10X5 with the "mean_pearson_corr" metric.