Commit Graph

398 Commits

Author SHA1 Message Date
Alexander Alekhin
b09a4a98d4 opencv: Use cv::AutoBuffer<>::data() 2018-07-04 19:11:29 +03:00
Alexander Alekhin
68c92908d5 Merge remote-tracking branch 'upstream/3.4' into merge-3.4 2018-05-14 15:17:35 +03:00
berak
9b0ef7bb17 ml: fix caching of internal state when changing the impl in KNearest 2018-05-11 12:20:17 +02:00
berak
fc5bba66af ml: refactor non-virtual methods 2018-04-24 13:23:27 +02:00
Alexander Alekhin
2385a5870e next(ml): eliminate dummy interface class ANN_MLP_ANNEAL 2018-04-10 18:09:54 +03:00
lopespt
65d816c3b5 Adds efficient sort algorithm to KNearest 2018-04-04 09:35:58 -03:00
Alexander Alekhin
4d0dd3e509 ml: apply CV_OVERRIDE/CV_FINAL 2018-03-28 18:43:27 +03:00
Vadim Pisarevsky
284e5231c5 Merge pull request #11171 from codingforfun:fix_11143 2018-03-28 12:47:37 +00:00
codingforfun
24e2e0d3f9 #11143 [FIX] Normalize node risk with sample weight sum
In case of regression trees, node risk is computed as sum of squared
error. To get a meaningfull value to compare with it needs to be
normalized to the number of samples in the node (or more generally to
the sum of sample weights in this node). Otherwise the sum of squared
error is highly dependend on the number of samples in the node and
comparision with `regressionAccuracy` parameter is not very meaningful.

After normalization `node_risk` means in fact sample variance for all
samples in the node, which makes much more sence and seams to be what
was originaly intended by the code given that node risk is later used as
a split termination criteria by
```
sqrt(node.node_risk) < params.getRegressionAccuracy()
```
2018-03-27 15:39:36 +02:00
Maksim Shabunin
92e9d4ec3a Fixed several issues detected by static analysis 2018-02-22 17:11:33 +03:00
luz.paz
e805a55a5b Misc. modules/ typos (cont.)
Found via `codespell`
2018-02-12 10:15:36 -05:00
Alexander Alekhin
00e43a9022 ml(ANN_MLP): ensure that train() call is always successful 2017-12-22 18:50:04 +00:00
Alexander Alekhin
289a8da39e ml: simplify interfaces of SimulatedAnnealingSolver 2017-12-22 16:35:48 +03:00
LaurentBerger
e43997dbb5 Calcerror uses now weighted samples (#10346)
* Calcerror uses now sample weights

* catree comment in #10319
2017-12-20 17:24:46 +03:00
Alexander Alekhin
aef3019152 ml: fix SimulatedAnnealingSolver interface 2017-12-15 21:44:32 +03:00
catree
0a439570a0 Move SimulatedAnnealingSolver::Impl in cpp file. Fix some typos. 2017-12-15 14:09:59 +01:00
LaurentBerger
7ad308ea47 Simulated Annealing for ANN_MLP training method (#10213)
* Simulated Annealing for ANN_MLP training method

* EXPECT_LT

* just to test new data

* manage RNG

* Try again

* Just run buildbot with new data

* try to understand

* Test layer

* New data- new test

* Force RNG in backprop

* Use Impl to avoid virtual method

* reset all weights

* try to solve ABI

* retry

* ABI solved?

* till problem with dynamic_cast

* Something is wrong

* Solved?

* disable backprop test

* remove ANN_MLP_ANNEALImpl

* Disable weight in varmap

* Add example for SimulatedAnnealing
2017-12-15 13:57:39 +03:00
Vadim Pisarevsky
614e254331 Merge pull request #10170 from LaurentBerger:Issue10166 2017-11-29 09:51:20 +00:00
LaurentBerger
a44573c43b Add ReLU and LeakyReLU activation function in ml module 2017-11-28 11:02:05 +01:00
LaurentBerger
606a5fd537 Try to solve issue 10166 2017-11-27 13:13:05 +01:00
LaurentBerger
b9cf65e905 Parallel version of calcError in statmodel 2017-11-09 21:29:06 +01:00
Vadim Pisarevsky
70c5d69640 Merge pull request #9310 from r2d3:svm_parallel 2017-09-18 09:32:41 +00:00
Vadim Pisarevsky
42cd061af0 Merge pull request #9581 from sovrasov:rtree_disable_cv 2017-09-13 13:20:52 +00:00
Maksim Shabunin
248e2c7d47 Fixed some issues found by static analysis 2017-09-08 12:22:12 +03:00
Vladislav Sovrasov
3eb9a655c7 ml: disable not implemented k-fold validation in RTrees 2017-09-07 15:49:46 +03:00
David Geldreich
7c334f45c6 trainAuto: uses parallel_for_ to dispatch all parameters combination to test 2017-08-08 13:27:02 +02:00
Rink Springer
3ce1bca670 Prevent crash when attempting to create training data without responses.
This is at least useful when using an SVM one-class linear classifier, so there are valid use cases.
2017-06-30 15:36:47 +02:00
Maksim Shabunin
a769d69a9d Fixed several issues found by static analysis 2017-06-28 18:06:18 +03:00
Maksim Shabunin
32d4af36e2 Fixing some static analysis issues 2017-06-27 14:30:26 +03:00
Alexander Alekhin
006966e629 trace: initial support for code trace 2017-06-26 17:07:13 +03:00
David Carlier
bacc210606 fixing segfaults occuring when launching those unit tests 2017-04-18 09:50:17 +01:00
Julian Tanke
f70cc29edb export SVM::trainAuto to python #7224 (#8373)
* export SVM::trainAuto to python #7224

* workaround for ABI compatibility of SVM::trainAuto

* add parameter comments to new SVM::trainAuto function

* Export ParamGrid member variables
2017-03-23 16:00:19 +03:00
Vadim Pisarevsky
e0ee2f769a Merge pull request #8116 from mrquorr:master 2017-03-02 11:07:23 +00:00
mrquorr
d8425d8881 finished for one sample
Finished with several samples support, need regression testing

Gave a more relevant name to function (getVotes)

Finished implicit implementation

Removed printf, finished regresion testing

Fixed conversion warning

Finished test for Rtrees

Fixed documentation

Initialized variable

Added doxygen documentation

Added parameter name
2017-02-28 11:14:33 -06:00
Alexander Alekhin
75533fcd06 Merge pull request #8098 from chrizandr:Tree_load_Wrapper
Add wrappers for load functions for DTrees and Boost classifiers
2017-01-30 17:25:23 +03:00
chrizandr
519fbdb8ab Wrappers for load methods of EM, LR, SVMSGD and Normal Bayes Classifier 2017-01-29 18:51:55 +05:30
chrizandr
d22df8c41f Add wrappers for load functions for DTrees and Boost classifiers 2017-01-29 15:55:38 +05:30
Vadim Pisarevsky
2bac66a181 Merge pull request #8004 from chrizandr:RTrees_load_wrapper 2017-01-24 11:55:42 +00:00
chrizandr
e3ec3566d9 RTrees: Add wrapper for RTrees_load method to enable loading trained RTrees 2017-01-19 17:16:45 +05:30
berak
89a740a62f Merge pull request #8019 from berak:patch-2
ml: fix small typo in lr.cpp (#8019)
2017-01-17 13:57:50 +03:00
Alexander Alekhin
23e53a32e5 Merge pull request #7950 from BadrinathS:firstbugfix-lr_opencv 2016-12-31 12:48:54 +00:00
BadrinathS
d1f727191f Resolving issue #7924 2016-12-31 16:37:51 +05:30
Vadim Pisarevsky
ae9f80c1a9 Merge pull request #7855 from logic1988:master 2016-12-19 13:45:51 +00:00
logic1988
de059567d7 Update inner_functions.cpp
Fix #4958 cv::ml::StatModel::calcError not working for responses of type CV_32S
2016-12-16 20:01:07 +03:00
Vladislav Sovrasov
d2e5bea1fb Disable training of RTrees when CVFolds > 0 2016-12-16 16:39:44 +03:00
Vadim Pisarevsky
40b870704e add 2 extra methods to ml::TrainData (#7169)
* expose 2 extra methods from ml::TrainData: getNames() and getVarSymbolFlags(). The first one returns text labels from CSV (if the data has been loaded from CSV); the second one returns a matrix of boolean values; its n-th element is 1 iff the corresponding column in the CSV uses symbolic names, not numbers.

* check that the dynamic_cast succeeds
2016-08-26 16:25:46 +04:00
LaurentBerger
b75bac7975 Solve Issue 7063
consequences of changes

accuracy test

Solve issue 7063
2016-08-11 10:56:50 +02:00
Vadim Pisarevsky
d7ee62f03b 1. fix warning from Xcode 7.x
2. fixed parsing of "cat[range_spec]ord[range_spec]" type specification string when using ml::TrainData::loadFromCSV(). Thanks to A. Kaehler for reporting it
2016-08-04 12:52:10 +03:00
Vadim Pisarevsky
b28194addc Merge pull request #6860 from IshankGulati:svm-error-message 2016-07-18 15:44:40 +00:00
Jose Luis Guardiola
2bb520e277 Fixed #6562: Incorrect CV_64F management in ANN 2016-05-20 13:16:59 +02:00
Maksim Shabunin
49b4af1e5c Merge pull request #6395 from Tauranis:master 2016-04-13 07:53:33 +00:00
Tauranis
edb6a0e889 Bug fix for MLP predict for small values to avoid nan responses. 2016-04-12 15:59:32 -03:00
Maksim Shabunin
8f6e102331 Merge pull request #6345 from eduardohenriquearnold:master 2016-04-12 15:22:32 +00:00
Maksim Shabunin
1307bb1d03 Merge pull request #6314 from mvukad:bugfix_dtreeswrite 2016-04-12 13:26:25 +00:00
mvukad
695e33b25b Fix missing format when writing Algorithm-based objects
Added a writeFormat() method to Algorithm which must be called by the
write() method of derived classes.
2016-04-07 13:49:42 -07:00
Vadim Pisarevsky
46d218bcec Merge pull request #6312 from KnockSoftware:split-float-epsilon 2016-04-03 23:10:10 +00:00
Eduardo Arnold
d046602ea4 Enforced DecisionFunction vector indexes to be saved on SVM save/load methods 2016-03-29 16:35:27 -03:00
Eugene Khvedchenya
ee92a36123 Added parallel implementation of compute_gradient method. 2016-03-29 11:09:54 +03:00
Evan Heidtmann
0712bccf52 Fix epsilon comparison when splitting 2016-03-28 14:16:32 -07:00
berak
4555f9ac25 svmsgd.cpp, fix #6248
add a cast to double, to make vs compilers happy
2016-03-14 08:02:09 +01:00
Vadim Pisarevsky
fbc221d334 Merge pull request #6096 from mnoskova:mn/SVMSGD_to_opencv3_0 2016-03-12 17:16:35 +00:00
Marina Noskova
53711ec29d Deleted default value for parameters in docs.
Added some asserts.
2016-02-25 19:12:54 +03:00
Marina Noskova
d484893839 Deleted functions makeTrainData() and makeTestData() in test_svmsgd.cpp.
Added function makeData() in test_svmsgd.cpp.
2016-02-25 16:57:03 +03:00
Marina Noskova
74c87a26a5 Delete function areClassesEmpty(). 2016-02-25 15:31:07 +03:00
Marina Noskova
f3c4a6aab8 Rename parameters lambda, gamma0 and c. 2016-02-24 13:22:07 +03:00
Marina Noskova
02cd8cf039 Deleted illegal type values. 2016-02-15 15:09:59 +03:00
Marina Noskova
ff54952769 Corrected spelling mistakes 2016-02-15 14:35:36 +03:00
Maksim Shabunin
54abb83c82 ml: moved getTestSamples implementation to src 2016-02-12 12:32:26 +03:00
Marina Noskova
5496dedd6a Fixed warnings. 2016-02-10 19:46:24 +03:00
Marina Noskova
617dd5db5b Fixed doc/opencv.bib 2016-02-10 17:57:36 +03:00
Marina Noskova
c522172367 Fixed small bug in SVMSGD::clear(). 2016-02-10 16:59:12 +03:00
Marina Noskova
05353a1492 Removed trailing whitespaces 2016-02-10 16:59:12 +03:00
Marina Noskova
41c0a38344 Fixed test samples for tests with different borders
Added new test (separating two points)
2016-02-10 16:59:12 +03:00
Marina Noskova
bfdca05f25 Added margin type, added tests with different scales of features.
Also fixed documentation, refactored sample.
2016-02-10 16:59:12 +03:00
Marina Noskova
acd74037b3 Increasing the dimension of features space in the SVMSGD::train function. 2016-02-10 16:59:11 +03:00
Marina Noskova
40bf97c6d1 Refactored SVMSGD class 2016-02-10 16:56:14 +03:00
joao.faro
a2f0963d66 SVMSGD class added 2016-02-10 16:53:15 +03:00
Alexander Alekhin
df89e76fb1 Merge pull request #5922 from DarwinsBuddy:fix_no_py_load_svm_bug 2016-01-14 14:44:27 +00:00
Alexander Alekhin
cb0b2bd1af Merge pull request #5965 from amroamroamro:fix_concentric_spheres 2016-01-14 12:09:25 +00:00
Amro
47cdb041f2 fix assignment of class labels
by using the indices from the sorted distance vector of pairs.
2016-01-14 02:56:34 +02:00
Amro
542f2b2e3c clean up code for train and predict methods of LR 2016-01-13 20:47:07 +02:00
Amro
e67178c696 fix LR predict with raw output in 2-class case
In case of binary class, pred_m is initialized as zero vector and later
returned when raw outputs are requested, but it is never filled.
2016-01-13 19:48:21 +02:00
Christoph Spörk
3f172731b2 added wrapped load function for python as suggested by gat3way 2016-01-11 10:59:15 +01:00
Christoph Spörk
6c8bc6a25b fixed ABI incompatibilities as proposed by alalek
related to issue 4969
fixes issue 5891
fixes issue 5922
2016-01-07 08:00:01 +01:00
Christoph Spörk
a7aa198b4c fixing issue #4969 of Itseez/opencv.
Someone forgot to wrap the load function for SVMs in the
corresponding ml python module. Fixed that.
2016-01-04 15:47:26 +01:00
berak
5afd0e211e ml: fix NormalBayesClassifier bulk prediction(#5911) 2016-01-04 11:47:08 +01:00
Philip
1ba2286c6d fix uninitialized matrix in EM::predict fixes #5443
fixes #5443
2015-12-23 11:19:48 +01:00
Ishank gulati
9636b5e821 reduce k_fold parameter 2015-12-22 15:47:11 +05:30
Dikay900
4f3b58d1e7 fix reshape call not being an inplace method
fixes #5853 #4740
2015-12-21 21:10:35 +01:00
Vadim Pisarevsky
e0395a79ec Merge pull request #5616 from afriesen:lr_predict 2015-12-11 12:22:56 +00:00
Vadim Pisarevsky
7a7b0bcfcb fixed the upper boundary when calling checkRange (thanks to alalek) 2015-12-10 20:17:17 +03:00
Deanna Hood
0d706f6796 Return uncompressed support vectors for getSupportVectors on linear SVM (Bug #4096) 2015-12-10 18:31:02 +03:00
Vadim Pisarevsky
544990e377 couple of small fixes in rtrees variable importance calculation 2015-12-10 18:13:54 +03:00
Vadim Pisarevsky
fda17273de applying patch by rxtsolar: https://github.com/Itseez/opencv/pull/5422 for the master branch (even though it's actually not that important here) 2015-12-10 18:10:27 +03:00
niederb
d8e3971e7f Fixed variable importance in rtrees 2015-12-10 18:09:15 +03:00
Tian Zhi
f0434d60b0 fixed the type inconsistent with document.
Document say probs will have CV_64F type. But in effect, it has CV_32F type.
http://docs.opencv.org/3.0.0/d1/dfb/classcv_1_1ml_1_1EM.html#a2ea7da92a75bc7a7d665c241f547b9b9
2015-11-12 02:54:18 +08:00
Abe Friesen
9c6ff4d955 - LogisticRegressionImpl::predict() was changed to return the predicted value and to only write to the OutputArray results if specified (no longer segfaults).
- Refactored batch and mini_batch training to use a common gradient computation function (removed duplicate code).
- Altered the cost computation so that NAN is not computed unnecessarily.
- Greatly simplified (and sped up) the code that appends a column of 1s to the data.
- Minor code cleanup.

Removed unused variables.

Added cast to float to remove warning
2015-11-02 17:49:06 -08:00
Alexander Alekhin
dfec99691b Merge pull request #5370 from berak:fix_svm_autoTrain 2015-10-26 10:47:29 +00:00
Alexander Alekhin
1648e9292c Merge pull request #5431 from MiguelAlgaba:em_one_cluster 2015-10-19 15:27:35 +00:00
Amro
13a0a37e63 fix randMVNormal in ML (#5469)
Fix the failed assertion by replacing the GEMM call.

Also random numbers are generated from normal distribution ~N(0,1),
instead of uniform distribution ~U(0,1).
2015-10-11 01:54:11 +03:00