2020-11
- rerun exerc05c to compare HMM results for I and I2 with the merged I+I2=II in exerc06
2020-10
- add exerc06
- as exerc05c but merging "I" and "I2" (into a "II" class)
- HMM part updated with a rerun using
double
for probabilities (ecoz2 v0.5.1) - add sequence length histograms
- add sequence classification rank vs. length scatter plot
- adjust exerc06/summary-parallel.py to use log scale for
M
- add some more HMM runs
- add MM exercise
2020-08
- add exerc05b, as exerc05 but only considering classes with at least 100 instances. Besides HMM and Naive Bayes, this also has the VQ based classification.
- add exerc05, as exerc1 but with the P = 20.
2020-06
- exerc02: run
c12n.plot.py
for classes A and F (test cases) - exerc02: For possible reference, exerc02/c12n/TRAIN/ with some similar plots but for training instances.
2020-05
-
rerun exerc02, resulting in an increase in average accuracy to 75% from 70%.
-
add exerc02, basically a rerun of exerc1 with the same base signal but with different train and test sets and also using new file organization is ecoz2 (based on a tt-list.csv)
-
remove data from version control to simplify things a bit
-
general update of the exerc01 exercise, see exerc01/README.md.
2020-03
- run VQ based classification. see exerc01/vq.md.
- HMM training parameter variations exerc01/README.md.
2020-03-14
- add exercise on MARS_20161221_000046_SongSession_16kHz_HPF5HzNorm_labels. Note: Direct run of the processing commands with similar parameters as in initial exercises, in particular, no model tuning at all.
2019-07-07
- Include the annotated selection data to make this repo more self-contained (except for the ECOZ2 executables).
- Point to the Oct 1, 2018 presentation.
2018-09-26
- Initial commit with complete exercises