Commit Graph

360 Commits

Author SHA1 Message Date
Gilles Louppe 44d81a0ec3 FIX: add except* in Splitter.init definitions 2015-02-23 08:19:10 +01:00
Arnaud Joly c1894b5223 FIX explicit initialization of normalizer 2014-11-28 18:58:58 +01:00
Arnaud Joly fd791ca68b ENH Release gil in feature importance 2014-11-28 18:33:26 +01:00
Arnaud Joly c95703a4b0 FIX unlikely pickling error of splitters 2014-11-26 11:03:37 +01:00
Arnaud Joly ab57964f74 COSMIT add Base prefix to DenseSplitter and DenseSplitter 2014-11-17 10:27:44 +01:00
Arnaud Joly 4f423d6e28 Remove constant print 2014-11-06 14:00:00 +01:00
Arnaud Joly bf9891683b Temporarily allows to set algorithm switching through an environment variable 2014-11-06 13:58:34 +01:00
Arnaud Joly c31d5653b3 wip benchmark 2014-11-06 13:58:34 +01:00
Arnaud Joly 9f3f5bb825 Extract non zero value extraction constant 2014-11-06 13:58:33 +01:00
Arnaud Joly 38183c835f ENH move utils near its use 2014-11-06 13:58:33 +01:00
Arnaud Joly 3047cd617d Revert previous version 2014-11-06 13:58:32 +01:00
Arnaud Joly 1c26cec286 ENH expand ternary operator 2014-11-06 13:58:32 +01:00
Arnaud Joly cb115118fc COSMIT simplify function call 2014-11-06 13:58:32 +01:00
Arnaud Joly cb9b741993 cosmit 2014-11-06 13:58:31 +01:00
Arnaud Joly 306924b526 ENH remove spurious code 2014-11-06 13:58:31 +01:00
Arnaud Joly 6cd9333596 FIX min_weight_leaf in best sparse splitter 2014-11-06 13:58:31 +01:00
Arnaud Joly 70226d05d8 FIX min_weight_fraction_split with random splitter 2014-11-06 13:58:30 +01:00
Arnaud Joly 62893e3ef4 ENH reduce number of parameters 2014-11-06 13:58:30 +01:00
Arnaud Joly 9dd87ad3da ENH while -> for loop 2014-11-06 13:58:29 +01:00
Arnaud Joly 0e86dfd5b0 Simplify call to extract_nnz making it a method 2014-11-06 13:58:29 +01:00
Arnaud Joly 7cb9a5cca1 Re-organize code dense splitter then sparse splitter 2014-11-06 13:58:29 +01:00
Arnaud Joly 0bc8a986d0 FIX+ENH add min_weight_fraction_split support for sparse splitter 2014-11-06 13:58:28 +01:00
Arnaud Joly 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
Lars Buitinck 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
jnothman 6d8ccbcb98 DOC correct shape of Tree.value 2014-08-14 19:09:21 +10:00
Arnaud Joly ea66557cdb ENH improve rand_int and rand_uniform 2014-07-16 16:03:43 +02:00
Lars Buitinck f7e95277b9 FIX potential overflow in _tree.safe_realloc
The multiplication with the sizeof the base type may overflow, esp.
on 32-bit machines with very large datasets, causing the function to
allocate less memory than was intended, without not raising an exception.
2014-07-15 13:49:17 +02:00
Noel Dawe 72b3da8302 tree: use min_weight_leaf internally while exposing min_weight_fraction_leaf 2014-07-02 11:47:32 +02:00
Noel Dawe 7e9c093585 tree: add min_weight_fraction_leaf 2014-07-02 11:47:31 +02:00
Joel Nothman 10507dcaa5 COSMIT factor out split data as struct 2014-06-22 23:55:38 -04:00
Lars Buitinck 65bb024f95 FIX MemoryError raising in trees (+test)
I got this wrong all along: Cython functions that don't return Python
objects need an "except" clause to actually propagate exceptions (spotted
by @GaelVaroquaux, @arjoly).

Simplified memory management and unified error messages by safe_realloc.
2014-05-05 19:44:31 +02:00
Arnaud Joly 9a91431db5 FIX raise memory error if constant_feature can't be realloc 2014-04-28 10:44:23 +02:00
Gilles Louppe efd610d801 ENH: remove unused variables 2014-04-13 20:33:00 +02:00
Gilles Louppe 7ce38f4405 ENH: better API for passing weighted_n_samples 2014-04-12 13:36:16 +02:00
Gilles Louppe 9f0828a044 ENH: renamed things too fast... 2014-04-12 13:28:12 +02:00
Gilles Louppe 45c0f046d0 ENH: rename normalizer -> weighted_n_samples 2014-04-12 13:25:35 +02:00
Gilles Louppe 57a743a157 ENH: rename Node attributes to avoid confusion 2014-04-12 11:06:57 +02:00
Gilles Louppe dcc6cc1496 ENH: declare weighted_n_samples 2014-04-12 10:47:55 +02:00
Gilles Louppe 698eafa4ed DOC: better doc for impurity_improvement 2014-04-12 10:38:42 +02:00
Gilles Louppe 61dd4b4be9 FIX: weight impurity improvement by the size of the node 2014-04-12 10:33:04 +02:00
Gilles Louppe 2633774569 PEP8 2014-04-10 20:29:14 +02:00
Gilles Louppe da95ad8a65 FIX: Set correct impurity values in BestFirstBuilder 2014-04-10 20:27:37 +02:00
cgohlke c95fcd0181 FIX MSVC compile error C2036: 'void *' : unknown size
We were inadvertently using a GCC extension (pointer arithmetic on
void *). Fixes #2963, #3016.
2014-03-30 11:18:30 +02:00
Arnaud Joly 2e5b210984 ENH rename CONSTANT_FEATURE_THRESHOLD to FEATURE_THRESHOLD + revert some pep8 fix for readability 2014-03-21 15:47:40 +01:00
Arnaud Joly 2c82c8adf0 FIX tons of flake8 2014-03-20 14:20:35 +01:00
Arnaud Joly 9d728f3f7d ENH stricter separation between splitter and criterion 2014-03-20 14:20:35 +01:00
Arnaud Joly 24db12ce77 ENH stop splitting for best first tree builder if pure node + change strict to unstrict inequality 2014-03-20 14:20:35 +01:00
Arnaud Joly 8534989c12 Document maximal depth of the tree 2014-03-20 14:20:35 +01:00
Arnaud Joly 203a598902 ENH add a dedicated constant for min_impurity_split 2014-03-20 14:20:35 +01:00
Arnaud Joly 39a2a76a2f ENH stricter separation between tree builder, splitter and criterion 2014-03-20 14:20:35 +01:00