lorenzo.volpi
/
QuaPy
forked from moreo/QuaPy
1
0
Fork 0
Code Pull Requests Activity
QuaPy/TODO.txt

22 lines
1.1 KiB
Plaintext
Raw Normal View History

svmperf wrapper added
2020-12-03 16:59:13 +01:00
Documentation with sphinx
QuaNet added, two examples of TextClassifiers added (CNN, LSTM)
2020-12-29 20:33:59 +01:00
Add quantification_report (akin to classification_report from sklearn) (?)
evaluation by artificial prevalence sampling added. New methods added. New util functions added to quapy.functional and quapy.utils
2020-12-10 19:04:33 +01:00
Add NAE, NRAE
Add "measures for evaluating ordinal"?
aggregative methods adapted. Explicit loss minimization methods (SVMQ, SVMKLD, ...) added and with support to binary or single-label. HDy added
2020-12-04 19:32:08 +01:00
Document methods with paper references
refactoring aggregative methods as methods that not only implement 'classify' and 'quantify', but that also implement 'aggregate' and that, by default, have a default implementation of 'quantify' as a pipeline of 'classify' and 'aggregate'; this helps speeding up evaluations A LOT, since the documents can be pre-classified and the samples are carried out across pre-classified values (labels, or posterior probabilities), and thus only aggregate is called many times within the artificial sampling protocol
2020-12-11 19:28:17 +01:00
The parallel training in svmperf seems not to work (not sure...)
QuaNet added, two examples of TextClassifiers added (CNN, LSTM)
2020-12-29 20:33:59 +01:00
In binary quantification (hp, kindle, imdb) we used F1 in the minority class (which in kindle and hp happens to be the
negative class). This is not covered in this new implementation, in which the binary case is not treated as such, but as
an instance of single-label with 2 labels. Check
Add classnames to LabelledCollection ?
Check the overhead in OneVsAll for SVMperf-based (?)
added Ensemble methods (methods ALL, ACC, Ptr, DS from Pérez-Gallego et al 2017 and 2019) and some UCI ML datasets used in those articles (only 5 datasets out of 32 they used)
2021-01-06 14:58:29 +01:00
Add HDy to QuaNet? if so, wrap HDy into OneVsAll in case the dataset is not binary.
Add datasets for topic.
Clarify whether QuaNet is an aggregative method or not.
Add medium swap method
Explore the hyperparameter "number of bins" in HDy
Implement HDy for single-label?
Rename EMQ to SLD ?
How many times is the system of equations for ACC and PACC not solved? How many times is it clipped? Do they sum up
adding tweet sent quant experiments
2021-01-11 18:31:12 +01:00
to one always?
refactor: methods requiring a val_split can now declare a default value in the __init__ method that will be used in case the fit method is called without specifying the val_split, which now is by default None in the fit, i.e., by default takes the value of the init, that is generally set to 0.4; some uci datasets added; ensembles can now be optimized for quantification, and can be trained on samples of smaller size
2021-01-22 18:01:51 +01:00
Parallelize the kFCV in ACC and PACC