9f45e72388
typo fix in comments
2019-06-20 17:02:14 -04:00
4df020df83
Typo fix in comments
2019-06-20 16:58:44 -04:00
41a4081f48
Typo fix in comments
2019-06-20 16:57:33 -04:00
96262c0c39
Error Fix: remove tqdm import
2019-06-19 17:36:04 -04:00
f33c97e720
BIG CHANGE: Remove tqdm as a package dependency. instead warn to install.
2019-06-19 17:15:37 -04:00
87b8fa15a3
Remove assertions that could trouble. fix a comment
2019-03-13 18:24:06 -04:00
95228b3e4d
MAJOR CHANGES: remove maxpred==label, multi-label support, simplified ordering
2019-02-26 16:55:04 -05:00
fc5cca124b
Fixed bug missing numpy prefix
2019-02-23 03:41:32 -05:00
237036b136
Version 0.0.8: supports full auto parallelization and many algorithm changes.
...
0.0.8 - Multiple major changes
- Finding label errors is no fully parallelized.
- prune_count_method parameter has been removed.
- estimate_confident_joint_from_probabilities now automatically calibrates confident joint to be a true joint estimate.
- Confident joint algorithm changed! When an example is found confidently as 2+ labels, choose class with max probability.
2019-02-17 16:09:32 -08:00
1c7d341a9a
MAJOR CHANGE: multiprocessed ordering, removed prune_count_method
2019-02-16 03:51:24 -05:00
3533b2afc6
Remove all references to prune_count_method and warnings
2019-02-16 00:20:16 -05:00
18637d17eb
Update comments to reflect that conf joint is calibrated now.
2019-02-15 23:06:25 -05:00
a712d18bd7
Replace cj with confident_joint
2019-02-15 22:56:21 -05:00
f44372cf31
Return array of False (isntead of None) when no label errors
2019-02-15 22:44:56 -05:00
e7c961a9cc
BIG CHANGE: Confident learning now uses multiprocessing for finding label errors.
2019-02-15 22:23:58 -05:00
3d4570d472
Removed reference to non-existent variable labels.
2019-02-15 19:43:01 -05:00
535c61a71d
Add method to estimate the full joint of noisy labels.
2019-02-15 19:20:07 -05:00
305eb61fc1
Add method to calibrate confident joint.
2019-02-15 19:13:55 -05:00
e7bc4276a0
MAJOR CHANGE: confident joint now uses max pred for collisions.
2019-02-15 13:54:57 -05:00
574346e7ab
Add environment bang header.
2019-02-14 19:27:16 -05:00
c65851685f
Avoid errors caused by letting s be a list. convert s to np array
2019-02-14 19:26:56 -05:00
66de6a42fc
Update docstsring to explain multi_label param
2019-02-14 18:35:04 -05:00
8b1897412e
MAJOR CHANGE - confident joint only selects each example once now.
2019-01-23 04:49:16 -05:00
6658e483c5
Add support for multi-label classification.
2019-01-21 20:59:56 -05:00
cbb3a81715
Add buffer to avoid precision issues with float.
2019-01-21 20:54:47 -05:00
81327c064f
0.0.7 - Major speed increases across the board. No major API changes.
2019-01-05 22:06:20 -08:00
70c8999d5c
avoid loop to handle end of file more efficiently.:
2019-01-04 14:50:45 -08:00
3619478bf3
Fixed errors. data_loader much faster. removed idx in indices.
2019-01-01 21:44:03 -08:00
be192e156f
Fixed error popping from empty stack
2019-01-01 21:18:36 -08:00
6ddcc7055b
Avoid idx in indices -- slow operation in _create_train
2019-01-01 21:09:20 -08:00
c8dee91f46
Initialize clf to None and do not __deepcopy__ when none.
2018-12-30 20:26:40 -08:00
d6a27ce212
Add support for copy.deepcopy
2018-12-30 19:28:20 -08:00
6c07d4c2da
Error fix. data_indices now correctly writes data for each index
2018-12-30 17:52:55 -08:00
6118bfd8d1
Error fix. Wrong file name.
2018-12-30 17:35:49 -08:00
1d41fffc21
Error fix. seperate write file from read file.
2018-12-30 17:33:07 -08:00
8662d8f153
Fixed bug. changed append with a+ to a
2018-12-30 17:25:37 -08:00
3d878b9d97
Add space around __newline__
2018-12-30 16:58:03 -08:00
c2c2843cd1
Fix error - replace newlines in text with __newline__
2018-12-30 16:56:35 -08:00
da9407f50d
Strip newline at end of first line read in data loader.
2018-12-30 15:53:00 -08:00
0b5bf3594c
Fixing error. Removing \n when writing data
2018-12-30 13:45:28 -08:00
f2295f6d73
Fix \n error by Replacing with __newline__
2018-12-30 13:25:32 -08:00
d4ca5fcdb7
Complete fefactor to use dataloader for large datasets in all methods.
2018-12-30 12:51:02 -08:00
79aac5e719
No longer require train_fn and allow specifying labels in constructor.
2018-12-28 16:21:36 -08:00
7763057014
Fixed bug. Wrong name used for variable.
2018-12-26 10:26:46 -08:00
adca853d50
Shortened number_label_errors to num_label_errors
2018-12-26 09:05:42 -08:00
e52a6f57aa
Added method to compute number of label errors
2018-12-26 09:04:48 -08:00
ca81c5fcf2
Creates copy of clf in cross validation to prevetn cross training between folds.
2018-12-25 22:40:29 -08:00
e1772c2eb6
fixed gpu compatibility issue
2018-12-25 22:32:19 -08:00
cd5ca048fc
fixed bug. fasttext superised kwargs now work when none are given
2018-12-24 22:35:15 -08:00
49af85e6ae
Fixed error that occurs when num2prune == len(s)
2018-12-24 13:03:53 -08:00
Curtis G. Northcutt