Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[REVIEW] Deprecating quantile-per-tree and removing three previously deprecated Random Forest parameters #3667

Merged
merged 4 commits into from
Mar 30, 2021
Merged

[REVIEW] Deprecating quantile-per-tree and removing three previously deprecated Random Forest parameters #3667

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2a7a9bf
Deprecating quantile-per-tree and removing three previously deprecate…
vinaydes Mar 29, 2021
53087b8
Updating copyright year
vinaydes Mar 29, 2021
4c95de7
Changing seed to random_state in RF test
vinaydes Mar 29, 2021
4bd491a
Changing seed to random_state in RF notebook
vinaydes Mar 30, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion docs/source/estimator_intro.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,7 +94,7 @@
"\n",
"model = cuRF( max_depth = max_depth, \n",
" n_estimators = n_estimators,\n",
" seed = 0 )\n",
" random_state = 0 )\n",
"\n",
"trained_RF = model.fit ( X_train, y_train )\n",
"\n",
Expand Down
2 changes: 1 addition & 1 deletion notebooks/random_forest_demo.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,7 @@
"cuml_model = curfc(n_estimators=40,\n",
" max_depth=16,\n",
" max_features=1.0,\n",
" seed=10)\n",
" random_state=10)\n",
"\n",
"cuml_model.fit(X_cudf_train, y_cudf_train)"
]
Expand Down
57 changes: 16 additions & 41 deletions python/cuml/ensemble/randomforest_common.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#
# Copyright (c) 2020, NVIDIA CORPORATION.
# Copyright (c) 2020-2021, NVIDIA CORPORATION.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -55,22 +55,17 @@ class BaseRandomForestModel(Base):

classes_ = CumlArrayDescriptor()

def __init__(self, *, split_criterion, seed=None,
n_streams=8, n_estimators=100,
max_depth=16, handle=None, max_features='auto',
n_bins=8, split_algo=1, bootstrap=True,
bootstrap_features=False,
verbose=False, min_rows_per_node=None,
min_samples_leaf=1, min_samples_split=2,
rows_sample=None, max_samples=1.0, max_leaves=-1,
accuracy_metric=None, dtype=None,
output_type=None,
min_weight_fraction_leaf=None, n_jobs=None,
max_leaf_nodes=None, min_impurity_decrease=0.0,
min_impurity_split=None, oob_score=None,
random_state=None, warm_start=None, class_weight=None,
quantile_per_tree=False, criterion=None,
use_experimental_backend=False, max_batch_size=128):
def __init__(self, *, split_criterion, n_streams=8, n_estimators=100,
max_depth=16, handle=None, max_features='auto', n_bins=8,
split_algo=1, bootstrap=True, bootstrap_features=False,
verbose=False, min_samples_leaf=1, min_samples_split=2,
max_samples=1.0, max_leaves=-1, accuracy_metric=None,
dtype=None, output_type=None, min_weight_fraction_leaf=None,
n_jobs=None, max_leaf_nodes=None, min_impurity_decrease=0.0,
min_impurity_split=None, oob_score=None, random_state=None,
warm_start=None, class_weight=None, quantile_per_tree=False,
criterion=None, use_experimental_backend=False,
max_batch_size=128):

sklearn_params = {"criterion": criterion,
"min_weight_fraction_leaf": min_weight_fraction_leaf,
Expand All @@ -89,37 +84,17 @@ class BaseRandomForestModel(Base):
"(https://docs.rapids.ai/api/cuml/nightly/"
"api.html#random-forest) for more information")

if seed is not None:
if random_state is None:
warnings.warn("Parameter 'seed' is deprecated and will be"
" removed in 0.17. Please use 'random_state'"
" instead. Setting 'random_state' as the"
" curent 'seed' value",
DeprecationWarning)
random_state = seed
else:
warnings.warn("Both 'seed' and 'random_state' parameters were"
" set. Using 'random_state' since 'seed' is"
" deprecated and will be removed in 0.17.",
DeprecationWarning)

if ((random_state is not None) and (n_streams != 1)):
warnings.warn("For reproducible results in Random Forest"
" Classifier or for almost reproducible results"
" in Random Forest Regressor, n_streams==1 is "
"recommended. If n_streams is > 1, results may vary "
"due to stream/thread timing differences, even when "
"random_state is set")
if min_rows_per_node is not None:
warnings.warn("The 'min_rows_per_node' parameter is deprecated "
"and will be removed in 0.18. Please use "
"'min_samples_leaf' parameter instead.")
min_samples_leaf = min_rows_per_node
if rows_sample is not None:
warnings.warn("The 'rows_sample' parameter is deprecated and will "
"be removed in 0.18. Please use 'max_samples' "
"parameter instead.")
max_samples = rows_sample
if quantile_per_tree:
warnings.warn("The 'quantile_per_tree' parameter is deprecated "
"and will be removed in 0.20 release. Instead use "
"higher number of global quantile bins.")
if handle is None:
handle = Handle(n_streams)

Expand Down
81 changes: 41 additions & 40 deletions python/cuml/ensemble/randomforestclassifier.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -129,40 +129,40 @@ class RandomForestClassifier(BaseRandomForestModel,
Implements a Random Forest classifier model which fits multiple decision
tree classifiers in an ensemble.

Note that the underlying algorithm for tree node splits differs from that
used in scikit-learn. By default, the cuML Random Forest uses a
histogram-based algorithms to determine splits, rather than an exact
count. You can tune the size of the histograms with the n_bins parameter.

.. note:: This is an early release of the cuML
Random Forest code. It contains a few known limitations:

* GPU-based inference is only supported if the model was trained
with 32-bit (float32) datatypes. CPU-based inference may be used
in this case as a slower fallback.
* Very deep / very wide models may exhaust available GPU memory.
Future versions of cuML will provide an alternative algorithm to
reduce memory consumption.
* While training the model for multi class classification problems,
using deep trees or `max_features=1.0` provides better performance.
.. note:: Note that the underlying algorithm for tree node splits differs
from that used in scikit-learn. By default, the cuML Random Forest uses a
histogram-based algorithm to determine splits, rather than an exact
count. You can tune the size of the histograms with the n_bins parameter.

**Known Limitations**: This is an early release of the cuML
Random Forest code. It contains a few known limitations:

* GPU-based inference is only supported if the model was trained
with 32-bit (float32) datatypes. CPU-based inference may be used
in this case as a slower fallback.
* Very deep / very wide models may exhaust available GPU memory.
Future versions of cuML will provide an alternative algorithm to
reduce memory consumption.
* While training the model for multi class classification problems,
using deep trees or `max_features=1.0` provides better performance.

Examples
--------
.. code-block:: python

import numpy as np
from cuml.ensemble import RandomForestClassifier as cuRFC
import numpy as np
from cuml.ensemble import RandomForestClassifier as cuRFC

X = np.random.normal(size=(10,4)).astype(np.float32)
y = np.asarray([0,1]*5, dtype=np.int32)
X = np.random.normal(size=(10,4)).astype(np.float32)
y = np.asarray([0,1]*5, dtype=np.int32)

cuml_model = cuRFC(max_features=1.0,
n_bins=8,
n_estimators=40)
cuml_model.fit(X,y)
cuml_predict = cuml_model.predict(X)
cuml_model = cuRFC(max_features=1.0,
n_bins=8,
n_estimators=40)
cuml_model.fit(X,y)
cuml_predict = cuml_model.predict(X)

print("Predicted labels : ", cuml_predict)
print("Predicted labels : ", cuml_predict)

Output:

Expand All @@ -180,7 +180,7 @@ class RandomForestClassifier(BaseRandomForestModel,
(default = 0)
split_algo : int (default = 1)
The algorithm to determine how nodes are split in the tree.
0 for HIST and 1 for GLOBAL_QUANTILE. HIST curently uses a slower
0 for HIST and 1 for GLOBAL_QUANTILE. HIST currently uses a slower
tree-building algorithm so GLOBAL_QUANTILE is recommended for most
cases.
bootstrap : boolean (default = True)
Expand Down Expand Up @@ -226,24 +226,25 @@ class RandomForestClassifier(BaseRandomForestModel,
Minimum decrease in impurity requried for
node to be spilt.
quantile_per_tree : boolean (default = False)
Whether quantile is computed for individal trees in RF.
Only relevant for GLOBAL_QUANTILE split_algo.
Whether quantile is computed for individual trees in RF.
Only relevant when `split_algo = GLOBAL_QUANTILE`.

.. deprecated:: 0.19
Parameter 'quantile_per_tree' is deprecated and will be removed in
subsequent release.
use_experimental_backend : boolean (default = False)
If set to true and following conditions are also met, experimental
decision tree training implementation would be used:
split_algo = 1 (GLOBAL_QUANTILE)
quantile_per_tree = false (No per tree quantile computation)
decision tree training implementation would be used only if
`split_algo = 1` (GLOBAL_QUANTILE) and `quantile_per_tree = False`
(No per tree quantile computation).
max_batch_size: int (default = 128)
Maximum number of nodes that can be processed in a given batch. This is
used only when 'use_experimental_backend' is true.
used only when 'use_experimental_backend' is true. Does not currently
fully guarantee the exact same results.
random_state : int (default = None)
Seed for the random number generator. Unseeded by default.
seed : int (default = None)
Seed for the random number generator. Unseeded by default.

.. deprecated:: 0.16
Parameter `seed` is deprecated and will be removed in 0.17. Please
use `random_state` instead
Seed for the random number generator. Unseeded by default. Does not
currently fully guarantee the exact same results. **Note: Parameter
`seed` is removed since release 0.19.**

handle : cuml.Handle
Specifies the cuml.handle that holds internal CUDA state for
Expand Down
49 changes: 23 additions & 26 deletions python/cuml/ensemble/randomforestregressor.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,21 +112,20 @@ class RandomForestRegressor(BaseRandomForestModel,
Implements a Random Forest regressor model which fits multiple decision
trees in an ensemble.

.. note:: that the underlying algorithm for tree node splits differs from
that used in scikit-learn. By default, the cuML Random Forest uses a
histogram-based algorithm to determine splits, rather than an exact
count. You can tune the size of the histograms with the n_bins
parameter.
.. note:: Note that the underlying algorithm for tree node splits differs
from that used in scikit-learn. By default, the cuML Random Forest uses a
histogram-based algorithm to determine splits, rather than an exact
count. You can tune the size of the histograms with the n_bins parameter.

**Known Limitations**: This is an early release of the cuML
Random Forest code. It contains a few known limitations:

* GPU-based inference is only supported if the model was trained
with 32-bit (float32) datatypes. CPU-based inference may be used
in this case as a slower fallback.
* Very deep / very wide models may exhaust available GPU memory.
Future versions of cuML will provide an alternative algorithm to
reduce memory consumption.
* GPU-based inference is only supported if the model was trained
with 32-bit (float32) datatypes. CPU-based inference may be used
in this case as a slower fallback.
* Very deep / very wide models may exhaust available GPU memory.
Future versions of cuML will provide an alternative algorithm to
reduce memory consumption.

Examples
--------
Expand All @@ -149,7 +148,7 @@ class RandomForestRegressor(BaseRandomForestModel,

Output:

.. code-block:: python
.. code-block:: none

MSE score of cuml : 0.1123437201231765

Expand All @@ -159,7 +158,7 @@ class RandomForestRegressor(BaseRandomForestModel,
Number of trees in the forest. (Default changed to 100 in cuML 0.11)
split_algo : int (default = 1)
The algorithm to determine how nodes are split in the tree.
0 for HIST and 1 for GLOBAL_QUANTILE. HIST curently uses a slower
0 for HIST and 1 for GLOBAL_QUANTILE. HIST currently uses a slower
tree-building algorithm so GLOBAL_QUANTILE is recommended for most
cases.
split_criterion : int (default = 2)
Expand Down Expand Up @@ -218,26 +217,24 @@ class RandomForestRegressor(BaseRandomForestModel,
for mean of abs error : 'mean_ae'
for mean square error' : 'mse'
quantile_per_tree : boolean (default = False)
Whether quantile is computed for individal trees in RF.
Only relevant for GLOBAL_QUANTILE split_algo.
Whether quantile is computed for individual trees in RF.
Only relevant when `split_algo = GLOBAL_QUANTILE`.

.. deprecated:: 0.19
Parameter 'quantile_per_tree' is deprecated and will be removed in
subsequent release.
use_experimental_backend : boolean (default = False)
If set to true and following conditions are also met, experimental
decision tree training implementation would be used:
split_algo = 1 (GLOBAL_QUANTILE)
quantile_per_tree = false (No per tree quantile computation)
decision tree training implementation would be used only if
`split_algo = 1` (GLOBAL_QUANTILE) and `quantile_per_tree = False`
(No per tree quantile computation).
max_batch_size: int (default = 128)
Maximum number of nodes that can be processed in a given batch. This is
used only when 'use_experimental_backend' is true.
random_state : int (default = None)
Seed for the random number generator. Unseeded by default. Does not
currently fully guarantee the exact same results.
seed : int (default = None)
Seed for the random number generator. Unseeded by default. Does not
currently fully guarantee the exact same results.

.. deprecated:: 0.16
Parameter `seed` is deprecated and will be removed in 0.17. Please
use `random_state` instead
currently fully guarantee the exact same results. **Note: Parameter
`seed` is removed since release 0.19.**

handle : cuml.Handle
Specifies the cuml.handle that holds internal CUDA state for
Expand Down
8 changes: 4 additions & 4 deletions python/cuml/test/test_random_forest.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -781,14 +781,14 @@ def test_rf_get_json(estimator_type, max_depth, n_estimators):
if estimator_type == 'classification':
cuml_model = curfc(max_features=1.0, max_samples=1.0,
n_bins=16, split_algo=0, split_criterion=0,
min_samples_leaf=2, seed=23707, n_streams=1,
min_samples_leaf=2, random_state=23707, n_streams=1,
n_estimators=n_estimators, max_leaves=-1,
max_depth=max_depth)
y = y.astype(np.int32)
elif estimator_type == 'regression':
cuml_model = curfr(max_features=1.0, max_samples=1.0,
n_bins=16, split_algo=0,
min_samples_leaf=2, seed=23707, n_streams=1,
min_samples_leaf=2, random_state=23707, n_streams=1,
n_estimators=n_estimators, max_leaves=-1,
max_depth=max_depth)
y = y.astype(np.float32)
Expand Down Expand Up @@ -862,7 +862,7 @@ def test_rf_instance_count(max_depth, n_estimators, use_experimental_backend):
X = X.astype(np.float32)
cuml_model = curfc(max_features=1.0, max_samples=1.0,
n_bins=16, split_algo=1, split_criterion=0,
min_samples_leaf=2, seed=23707, n_streams=1,
min_samples_leaf=2, random_state=23707, n_streams=1,
n_estimators=n_estimators, max_leaves=-1,
max_depth=max_depth,
use_experimental_backend=use_experimental_backend)
Expand Down Expand Up @@ -1031,7 +1031,7 @@ def test_rf_regression_with_identical_labels(split_criterion,
# Degenerate case: all labels are identical.
# RF Regressor must not create any split. It must yield an empty tree
# with only the root node.
clf = curfr(max_features=1.0, rows_sample=1.0, n_bins=5, split_algo=1,
clf = curfr(max_features=1.0, max_samples=1.0, n_bins=5, split_algo=1,
bootstrap=False, split_criterion=split_criterion,
min_samples_leaf=1, min_samples_split=2, random_state=0,
n_streams=1, n_estimators=1, max_depth=1,
Expand Down