min_impurity_split parameter of GradientBoostingRegressor is not used #9514
Comments
|
Hey @CyrilLeMat, you're totally right about how For example, from sklearn.ensemble import GradientBoostingRegressor
clf = GradientBoostingRegressor(min_impurity_split=-0.1)Doesn't raise an error, but from sklearn.ensemble import GradientBoostingRegressor
clf = GradientBoostingRegressor(min_impurity_split=-0.1)
# Make up some training data
X = [[1, 2, 3], [4, 5, 6]]
y = [0, 1]
# Fit classifier
clf.fit(X, y)raises |
|
Hi, GradientBoostingRegressor(alpha=0.9, criterion='friedman_mse', init=None,
learning_rate=0.1, loss='ls', max_depth=3, max_features=None,
max_leaf_nodes=None, min_impurity_split=-0.1,
min_samples_leaf=1, min_samples_split=2,
min_weight_fraction_leaf=0.0, n_estimators=100,
presort='auto', random_state=None, subsample=1.0, verbose=0,
warm_start=False) |
|
do you then call fit?
…On 10 Aug 2017 5:51 pm, "CyrilLeMat" ***@***.***> wrote:
Hi,
thanks for your answer,
However your exact code don't raise any error on my side.
it prints:
'''python
GradientBoostingRegressor(alpha=0.9, criterion='friedman_mse', init=None,
learning_rate=0.1, loss='ls', max_depth=3, max_features=None,
max_leaf_nodes=None, min_impurity_split=-0.1,
min_samples_leaf=1, min_samples_split=2,
min_weight_fraction_leaf=0.0, n_estimators=100,
presort='auto', random_state=None, subsample=1.0, verbose=0,
warm_start=False)
'''
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#9514 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AAEz6357C1NwqkJJTQ2iOfT2FtbF0BVKks5sWraSgaJpZM4OyUt9>
.
|
|
Hi! yes I did, I found the problem on my package : When the DecisionTreeRegressor are created, there is no parameter 'min_impurity_split =self. min_impurity_split' something new with the 0.18.2? (edit : sorry it's min_impurity_split, not min_samples_split) |
|
we can see here that the min_impurity_split is not used. however it seems to be back on the 0.19: |
|
Good catch, I should have specified that I'm using |
|
@CyrilLeMat but if there is no parameter of that name, you get an error, right? |
|
I just tried from sklearn.ensemble import GradientBoostingRegressor
clf = GradientBoostingRegressor(min_impurity_split=-0.1)
# Make up some training data
X = [[1, 2, 3], [4, 5, 6]]
y = [0, 1]
# Fit classifier
clf.fit(X, y)with v0.18.2 and it doesn't throw an error. There is a |
|
Is it passed correctly in v0.19? Looks like it is, to me.
…On 12 August 2017 at 09:29, James Bourbeau ***@***.***> wrote:
I just tried
from sklearn.ensemble import GradientBoostingRegressor
clf = GradientBoostingRegressor(min_impurity_split=-0.1)# Make up some training data
X = [[1, 2, 3], [4, 5, 6]]
y = [0, 1]# Fit classifier
clf.fit(X, y)
with v0.18.2 and it doesn't throw an error. There is a min_impurity_split
parameter for GradientBoostingRegressor in this version. However, during
fitting, the min_impurity_split value from GradientBoostingRegressor is
not passed to the DecisionTreeRegressor that is induced on the residuals
(where the validation of min_impurity_split occurs). So the default value
for min_impurity_split (1e-7) is used instead.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#9514 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AAEz66BxL0GW-ayIQsSXcubYNUBo3qUzks5sXOPegaJpZM4OyUt9>
.
|
|
Yeah, it's passed correctly in v0.19 |
Description
min_impurity_split parameter of GradientBoostingRegressor is not used
Steps/Code to Reproduce
Example:
Expected Results
This example should raise an error
Actual Results
No errors
Versions
Darwin-16.5.0-x86_64-i386-64bit
Python 3.6.1 (v3.6.1:69c0db5050, Mar 21 2017, 01:21:04)
[GCC 4.2.1 (Apple Inc. build 5666) (dot 3)]
NumPy 1.13.1
SciPy 0.19.0
Scikit-Learn 0.18.2
The text was updated successfully, but these errors were encountered: