 Ryan Curtin
|
7b28dce43d
|
Fix incremental single-point training.
|
2015-09-26 04:17:01 +00:00 |
|
|
Ryan Curtin
|
e59d52d95a
|
Refactor for new NBC API. Not yet tested.
|
2015-09-25 23:35:28 +00:00 |
|
|
Ryan Curtin
|
0ae0b1a9a4
|
Refactor tests for changed NBC API.
No new tests for Train() or empty constructor yet.
|
2015-09-25 22:53:31 +00:00 |
|
|
Ryan Curtin
|
4a469b2733
|
Refactor to add Train() and empty constructor.
|
2015-09-25 22:53:20 +00:00 |
|
|
Ryan Curtin
|
5b61871334
|
Include std::vector serialization support.
|
2015-09-25 22:43:25 +00:00 |
|
|
Ryan Curtin
|
075383bd53
|
Fix AllkNN serialization test.
|
2015-09-25 22:43:13 +00:00 |
|
|
Ryan Curtin
|
5cfb713309
|
Add Serialize().
|
2015-09-25 21:33:14 +00:00 |
|
|
Ryan Curtin
|
14e794d006
|
Add empty Serialize() methods to initialization policies.
|
2015-09-25 21:33:13 +00:00 |
|
|
Ryan Curtin
|
df2dac2a7e
|
Add unfinished serialization test.
|
2015-09-25 17:28:04 -04:00 |
|
|
Ryan Curtin
|
2fa4ba2ca3
|
Test empty NeighborSearch objects and Train().
|
2015-09-25 17:20:17 -04:00 |
|
|
Ryan Curtin
|
60c81702c4
|
Add Train() functions to NeighborSearch for consistency.
This will make serialization much easier, and also adds more flexibility to the
class.
|
2015-09-25 17:18:34 -04:00 |
|
|
Ryan Curtin
|
fa45524053
|
Handle empty datasets.
|
2015-09-25 16:28:40 -04:00 |
|
|
Ryan Curtin
|
7bd9c956ee
|
Fix unused parameter warning.
|
2015-09-25 11:40:20 -04:00 |
|
|
Ryan Curtin
|
7e682d53a3
|
Deal with newer Boost versions nicely.
|
2015-09-25 08:51:27 -04:00 |
|
|
Ryan Curtin
|
3ba7a678c1
|
When the user requests single-tree, actually do it.
|
2015-09-17 14:35:08 -04:00 |
|
|
Ryan Curtin
|
9295469715
|
Fix bugs in logistic_regression.
|
2015-09-16 23:04:24 +00:00 |
|
|
Ryan Curtin
|
273401efeb
|
Update documentation for logistic_regression.
|
2015-09-16 22:05:07 +00:00 |
|
|
Ryan Curtin
|
1eec7b7b95
|
Note the changes that have been made.
|
2015-09-16 21:44:18 +00:00 |
|
|
Ryan Curtin
|
d6e9b1be05
|
Refactor main executable. Still needs documentation.
|
2015-09-16 21:29:47 +00:00 |
|
|
Ryan Curtin
|
14d8551865
|
Add tests for new LogisticRegression functionality.
|
2015-09-16 21:29:36 +00:00 |
|
|
Ryan Curtin
|
f379ee2dd5
|
Add another Train(); remove templated constructors.
C++ language restrictions prevent the templated constructors from being usable.
|
2015-09-16 21:29:10 +00:00 |
|
|
Ryan Curtin
|
e67787e336
|
Refactor logistic_regression to allow persistent models.
|
2015-09-16 05:48:04 +00:00 |
|
|
Ryan Curtin
|
b34bb0f90d
|
Add and implement Train() methods.
|
2015-09-16 03:30:45 +00:00 |
|
|
Ryan Curtin
|
600b2f5d52
|
Refactor so that OptimizerType is only needed sometimes.
OptimizerType isn't relevant to an already-trained model; so it should only be specified during training.
|
2015-09-16 02:50:12 +00:00 |
|
|
Ryan Curtin
|
bbe9cd1615
|
Refactor LogisticRegressionFunction to work with arbitrary matrix types.
|
2015-09-15 02:59:12 +00:00 |
|
|
Ryan Curtin
|
ebfe0c0b9d
|
Wow, this was an incredibly stupid bug.
|
2015-09-15 02:59:12 +00:00 |
|
|
Marcus Edel
|
754ff98bb9
|
Build the convolution test.
|
2015-09-12 14:36:47 +07:00 |
|
|
Ryan Curtin
|
763f54cf37
|
Add Serialize() to LogisticRegression.
|
2015-09-11 16:05:26 +00:00 |
|
|
Ryan Curtin
|
aef36197cd
|
Refactor perceptron program to load/save models.
Now you can save perceptrons, which is a big functionality boost. Also the documentation has been redone.
|
2015-09-11 15:19:34 +00:00 |
|
|
Ryan Curtin
|
a33bc45442
|
Add non-training constructor and write Serialize().
|
2015-09-09 02:08:20 +00:00 |
|
|
ryan
|
a4d2dc275f
|
Update documentation.
|
2015-09-08 10:21:44 -04:00 |
|
|
ryan
|
ee732cae82
|
Allow .tsv extension too.
|
2015-09-08 09:53:44 -04:00 |
|
|
Ryan Curtin
|
14e1f4fb7c
|
Fix Adaboost executable for Perceptron API change.
|
2015-09-08 13:52:49 +00:00 |
|
|
Ryan Curtin
|
4bcc6e6409
|
Merge pull request #450 from thirdwing/master
load malformed CSVs correctly
|
2015-09-08 13:34:14 +00:00 |
|
|
Ryan Curtin
|
16cb76a12b
|
Require specification of number of classes.
|
2015-09-08 13:04:58 +00:00 |
|
|
Qiang Kou
|
d6a5f7058a
|
fix #449
|
2015-09-04 17:46:20 -04:00 |
|
|
Ryan Curtin
|
980b66c059
|
Rename iter -> maxIterations.
|
2015-09-04 18:11:06 +00:00 |
|
|
Ryan Curtin
|
2bd5e7889d
|
Refactor to make instance weights optional.
|
2015-09-04 18:06:49 +00:00 |
|
|
Ryan Curtin
|
8ace79ecba
|
Rename weightVectors to weights and simplify API.
|
2015-09-04 17:55:01 +00:00 |
|
|
Ryan Curtin
|
f3bd5e8853
|
Fix bug introduced in previous commits.
|
2015-09-04 17:30:58 +00:00 |
|
|
Ryan Curtin
|
a96135093f
|
Split biases into separate vector (speedup).
|
2015-09-04 16:35:33 +00:00 |
|
|
Ryan Curtin
|
105a30fbc8
|
Simplify expression.
|
2015-09-04 16:28:29 +00:00 |
|
|
Ryan Curtin
|
19a4620fc4
|
Refactor weight vectors to be column-major.
|
2015-09-04 16:28:00 +00:00 |
|
|
Ryan Curtin
|
424383cb02
|
Refactor perceptron to not modify input dataset.
Minimize internally-held variables.
|
2015-09-04 15:50:29 +00:00 |
|
|
Ryan Curtin
|
80feca52c7
|
Add Serialize() (unimplemented) to the perceptron.
|
2015-09-04 15:50:29 +00:00 |
|
|
Marcus Edel
|
f5893d5d19
|
Use static weights for the network decreasing error test.
|
2015-09-04 15:26:35 +02:00 |
|
|
Marcus Edel
|
ef29032111
|
Fix 'no known conversion from arma::Mat<double> to arma::Cube<double>' error that occurred on some systems. Actually the build should be fine even without the change, but I guess some systems like to draw our attention on some 'incredibly long' templates in the error log.
|
2015-09-04 14:46:59 +02:00 |
|
|
Marcus Edel
|
952cc39ac1
|
Use the logistic function and the RMSProp optimizer because the combination tends to be more stable and in most cases it converges faster using the specified architecture.
|
2015-09-04 13:58:44 +02:00 |
|
|
Marcus Edel
|
df9446c5a3
|
Comment out the convolution test.
|
2015-09-03 14:43:28 +02:00 |
|
|
Marcus Edel
|
c55ce61bb4
|
Widen convolutional test tolerance.
|
2015-09-03 14:40:59 +02:00 |
|