44d81a0ec3
FIX: add except* in Splitter.init definitions
2015-02-23 08:19:10 +01:00
c03c01a504
ENH Bring sparse input support to tree-based methods
...
Author: Arnaud Joly <arnaud.v.joly@gmail.com>
Fares Hedayati <fares.hedayati@gmail.com>
2014-11-06 13:58:28 +01:00
5335539d95
FIX propagate MemoryError from Tree._resize
...
Fixes #3684 . "except *" declaration was ignored because it was missing
from the .pxd file.
2014-09-22 14:22:46 +02:00
72b3da8302
tree: use min_weight_leaf internally while exposing min_weight_fraction_leaf
2014-07-02 11:47:32 +02:00
7e9c093585
tree: add min_weight_fraction_leaf
2014-07-02 11:47:31 +02:00
10507dcaa5
COSMIT factor out split data as struct
2014-06-22 23:55:38 -04:00
7ce38f4405
ENH: better API for passing weighted_n_samples
2014-04-12 13:36:16 +02:00
45c0f046d0
ENH: rename normalizer -> weighted_n_samples
2014-04-12 13:25:35 +02:00
57a743a157
ENH: rename Node attributes to avoid confusion
2014-04-12 11:06:57 +02:00
698eafa4ed
DOC: better doc for impurity_improvement
2014-04-12 10:38:42 +02:00
61dd4b4be9
FIX: weight impurity improvement by the size of the node
2014-04-12 10:33:04 +02:00
d40eeb0823
DOC: wordings
2014-03-22 17:41:33 +01:00
d774953ed8
DOC missing words
2014-03-21 15:53:01 +01:00
a4680b1268
DOC document internal tree class
2014-03-21 15:34:31 +01:00
3a6b8340d6
Add space to be consistent with _tree.pyx
2014-03-21 14:38:54 +01:00
5839ddaca4
DOC document max_depth of TreeBuilder
2014-03-20 14:20:35 +01:00
2c82c8adf0
FIX tons of flake8
2014-03-20 14:20:35 +01:00
9d728f3f7d
ENH stricter separation between splitter and criterion
2014-03-20 14:20:35 +01:00
39a2a76a2f
ENH stricter separation between tree builder, splitter and criterion
2014-03-20 14:20:35 +01:00
8f6b75375a
ENH for the tree structure max_depth is part of the inner structure
2014-03-20 14:20:35 +01:00
07d6e8d926
ENH clean public tree strucure
2014-03-20 14:20:34 +01:00
e22045a670
ENH avoid serializing the splitter & criterion + avoid storing tree parameter in the tree structure
2014-03-20 14:20:34 +01:00
f434c8cb84
Clean tree builder interface
2014-03-20 14:20:34 +01:00
57f73f16c5
DOC more comments about features and constant_features
2014-03-10 14:00:33 +01:00
c49294c392
DOC verbose documentation of features and constant_features
2014-03-10 12:17:52 +01:00
3e123a1b7f
WIP add constant_features array
2014-03-10 12:16:09 +01:00
abaf100a65
WIP add n_constant_features argument
2014-03-10 12:15:58 +01:00
4ebf883c38
FIX: Weight impurity decreases with weighted_n_samples
2014-02-06 19:01:32 +01:00
60e32899ef
Fixed documentation on sklearn/tree
2014-01-29 13:32:33 -02:00
7c33502c2e
ENH faster heapsort in trees
...
Uses the heapsort version that I'm familiar with: linear-time heapify
followed by n delete-min operations. Also cache-friendlier by copying
active features into a temporary array.
Random forest training time on covertype:
now 34.4123s
before 60.1179s
2014-01-25 15:13:01 +01:00
87194f77cb
Remove 'locked', add comments
2014-01-25 22:10:18 +11:00
86fbebdadf
FIX dtype refcount; block resizing after building
2014-01-24 09:30:06 +11:00
e53263988a
FIX remove reference cycles from Tree
2014-01-24 07:54:01 +11:00
e7e7133912
ENH/FIX Change Tree underlying data structure
...
* avoid segfault (#2726 ) by setting base on numpy arrays
* A struct array for tree nodes and values array (size indeterminate at
compile time) are the only underlying structure, so each node is locally
grouped memory, and joblib dumps will save only two numpy files per
tree.
* predict() uses the value array's take method, reducing code repetition
2014-01-16 09:32:16 +11:00
7fc52ffdb4
BUG fix unchecked mallocs in trees
...
All mallocs should be checked now. Some notes for the next person to touch
this code, or do anything with malloc in Cython code:
* with gil: raise MemoryError() didn't work as expected, as a surrounding
nogil block catches and logs the exception, then proceeds as if nothing
happened.
* There were several opportunities for double free in the code. This is
handled by letting the destructor handle the freeing.
* Similarly, constructors failed to set pointers to NULL so __dealloc__
would try to free unitialized pointers.
* Out of memory might leave trees in an inconsistent state, but will not
cause a double free. A subsequent successful fit should produce a
consistent state, although I didn't test this.
* Replaced a free/calloc with a realloc; realloc(NULL, n) = malloc(n).
* free(NULL) is a harmless no-op per ISO C89.
Also performed an optimization: the pre-sort splitter was using size_t for
an array of booleans. unsigned char is faster and smaller.
Finally, fixed some cosmetic issues (long lines).
2014-01-11 15:44:44 +01:00
ce0d52ea05
COSMITS + FIX to impurity_improvement
2013-11-29 14:02:21 +01:00
fda6211367
pass impurity to ``impurity_improvement`` - now it is correct so that arbitrary Criteria can be used in best-first search.
2013-11-29 10:56:57 +01:00
cde390f2c3
remove split_impurity from node_split signature
2013-11-22 17:47:21 +01:00
5847be48a5
removed Impurity struct, Criterion.child_impurity now returns impurity_left and impurity_right (not total), fix: moved pos >= end check to if branch
2013-11-22 11:28:33 +01:00
4b45668212
Best-first instead of branch heuristic - identical score as GBM on covertype
2013-11-21 13:43:54 +01:00
13e43d202b
Implement GBM's best-first heuristic tree growing procedure .
2013-11-20 21:10:16 +01:00
5f09b5a419
tree code now supports both c-style and fortran inputs
2013-11-03 14:19:49 +02:00
e9cea77bd8
PriorityQueue for best-first tree growing
...
max_leaf_nodes instead of complete parameter
2013-11-03 09:40:24 +02:00
1192fe64e5
Tree: compute partition specific impurity and total impurity in criterion.children_impurity. Pass partition impurity to stack to avoid re-computation (saves some runtime).
...
New tree parameter: complete; specifies whether complete binary trees are grows or if a greedy branch of max_depth is grown with at most max_depth + 1 leafs.
2013-11-02 01:04:23 +02:00
f859b7d3fa
Release the GIL at tree building time
2013-10-16 16:26:57 +02:00
6a5a65dede
ENH: use npy_int32 type
2013-10-09 13:36:02 +02:00
b9384c98cb
ENH: make node_reset/split/value gil-free
2013-10-09 13:35:52 +02:00
dc36bf3dc0
FIX: random number generator
...
FIX: address @larsmans + @ogrisel comments
ENH: add INT32_t type
ENH: rename INT32_t to UINT32_t
FIX: modulo RAND_R_MAX+1
FIX: broken tests
2013-09-22 19:35:18 +02:00
69f49b1532
FIX: release reference to X in _tree.splitter
2013-08-13 22:58:10 +02:00
20f5985d68
DOC: some documentation for the Tree Cython structure
2013-07-25 09:40:26 +02:00
Gilles Louppe