[GPU-Plugin] Change GPU plugin to use tree_method parameter, bump cmake version to 3.5 for GPU plugin, add compute architecture 3.5, remove unused cmake files (#2455)

2017-06-29 16:19:45 +12:00
parent 88488fdbb9
commit 48f3003302
8 changed files with 168 additions and 835 deletions
--- a/plugin/updater_gpu/README.md
+++ b/plugin/updater_gpu/README.md
@@ -1,16 +1,16 @@
 # CUDA Accelerated Tree Construction Algorithms
 This plugin adds GPU accelerated tree construction algorithms to XGBoost.
 ## Usage
-Specify the 'updater' parameter as one of the following algorithms. 
+Specify the 'tree_method' parameter as one of the following algorithms. 

 ### Algorithms
-| updater | Description |
+| tree_method | Description |
 | --- | --- |
-grow_gpu | The standard XGBoost tree construction algorithm. Performs exact search for splits. Slower and uses considerably more memory than 'grow_gpu_hist' |
-grow_gpu_hist | Equivalent to the XGBoost fast histogram algorithm. Faster and uses considerably less memory. Splits may be less accurate. |
+gpu_exact | The standard XGBoost tree construction algorithm. Performs exact search for splits. Slower and uses considerably more memory than 'gpu_hist' |
+gpu_hist | Equivalent to the XGBoost fast histogram algorithm. Faster and uses considerably less memory. Splits may be less accurate. |

 ### Supported parameters 
-| parameter | grow_gpu | grow_gpu_hist |
+| parameter | gpu_exact | gpu_hist |
 | --- | --- | --- |
 subsample | &#10004; | &#10004; |
 colsample_bytree | &#10004; | &#10004;|
@@ -29,7 +29,7 @@ Python example:
 ```python
 param['gpu_id'] = 1
 param['max_bin'] = 16
-param['updater'] = 'grow_gpu_hist'
+param['tree_method'] = 'gpu_hist'
 ```
 ## Benchmarks
 To run benchmarks on synthetic data for binary classification:
@@ -39,18 +39,18 @@ $ python benchmark/benchmark.py

 Training time time on 1000000 rows x 50 columns with 500 boosting iterations on i7-6700K CPU @ 4.00GHz and Pascal Titan X.

-| Updater | Time (s) |
+| tree_method | Time (s) |
 | --- | --- |
-| grow_gpu_hist | 11.09 |
-| grow_fast_histmaker (histogram XGBoost - CPU) | 41.75 |
-| grow_gpu | 193.90 |
-| grow_colmaker (standard XGBoost - CPU) | 720.12 |
+| gpu_hist | 11.09 |
+| hist (histogram XGBoost - CPU) | 41.75 |
+| gpu_exact | 193.90 |
+| exact (standard XGBoost - CPU) | 720.12 |


-[See here](http://dmlc.ml/2016/12/14/GPU-accelerated-xgboost.html) for additional performance benchmarks of the 'grow_gpu' updater.
+[See here](http://dmlc.ml/2016/12/14/GPU-accelerated-xgboost.html) for additional performance benchmarks of the 'gpu_exact' tree_method.

 ## Test
-To run tests:
+To run tests:Will
 ```bash
 $ python -m nose test/python/
 ```
@@ -122,6 +122,13 @@ $ make PLUGIN_UPDATER_GPU=ON GTEST_PATH=${CACHE_PREFIX} test
 ```

 ## Changelog
+##### 2017/6/26
+
+* Change API to use tree_method parameter
+* Increase required cmake version to 3.5
+* Add compute arch 3.5 to default archs
+* Set default n_gpus to 1
+
 ##### 2017/6/5

 * Multi-GPU support for histogram method using NVIDIA NCCL.
--- a/plugin/updater_gpu/benchmark/benchmark.py
+++ b/plugin/updater_gpu/benchmark/benchmark.py
@@ -14,19 +14,18 @@ def run_benchmark(args, gpu_algorithm, cpu_algorithm):
    dtrain = xgb.DMatrix(X, y)

    param = {'objective': 'binary:logistic',
-             'tree_method': 'exact',
             'max_depth': 6,
             'silent': 1,
             'eval_metric': 'auc'}

-    param['updater'] = gpu_algorithm
-    print("Training with '%s'" % param['updater'])
+    param['tree_method'] = gpu_algorithm
+    print("Training with '%s'" % param['tree_method'])
    tmp = time.time()
    xgb.train(param, dtrain, args.iterations)
    print ("Time: %s seconds" % (str(time.time() - tmp)))

-    param['updater'] = cpu_algorithm
-    print("Training with '%s'" % param['updater'])
+    param['tree_method'] = cpu_algorithm
+    print("Training with '%s'" % param['tree_method'])
    tmp = time.time()
    xgb.train(param, dtrain, args.iterations)
    print ("Time: %s seconds" % (str(time.time() - tmp)))
@@ -34,17 +33,17 @@ def run_benchmark(args, gpu_algorithm, cpu_algorithm):


 parser = argparse.ArgumentParser()
-parser.add_argument('--algorithm', choices=['all', 'grow_gpu', 'grow_gpu_hist'], required=True)
+parser.add_argument('--algorithm', choices=['all', 'gpu_exact', 'gpu_hist'], default='all')
 parser.add_argument('--rows',type=int,default=1000000)
 parser.add_argument('--columns',type=int,default=50)
 parser.add_argument('--iterations',type=int,default=500)
 args = parser.parse_args()

-if 'grow_gpu_hist' in args.algorithm:
-    run_benchmark(args, args.algorithm, 'grow_fast_histmaker')
-if 'grow_gpu' in args.algorithm:
-    run_benchmark(args, args.algorithm, 'grow_colmaker')
+if 'gpu_hist' in args.algorithm:
+    run_benchmark(args, args.algorithm, 'hist')
+if 'gpu_exact' in args.algorithm:
+    run_benchmark(args, args.algorithm, 'exact')
 if 'all' in args.algorithm:
-    run_benchmark(args, 'grow_gpu', 'grow_colmaker')
-    run_benchmark(args, 'grow_gpu_hist', 'grow_fast_histmaker')
+    run_benchmark(args, 'gpu_exact', 'exact')
+    run_benchmark(args, 'gpu_hist', 'hist')

--- a/plugin/updater_gpu/test/python/test.py
+++ b/plugin/updater_gpu/test/python/test.py
@@ -35,7 +35,7 @@ class TestGPU(unittest.TestCase):
                    'objective': 'binary:logistic',
                    'eval_metric': 'auc'}
        ag_param2 = {'max_depth': 2,
-                     'updater': 'grow_gpu',
+                     'tree_method': 'gpu_exact',
                     'eta': 1,
                     'silent': 1,
                     'objective': 'binary:logistic',
@@ -59,7 +59,7 @@ class TestGPU(unittest.TestCase):
        dtest = xgb.DMatrix(X_test, y_test)

        param = {'objective': 'binary:logistic',
-                 'updater': 'grow_gpu',
+                 'tree_method': 'gpu_exact',
                 'max_depth': 3,
                 'eval_metric': 'auc'}
        res = {}
@@ -75,7 +75,7 @@ class TestGPU(unittest.TestCase):
        dtrain2 = xgb.DMatrix(X2, label=y2)

        param = {'objective': 'binary:logistic',
-                 'updater': 'grow_gpu',
+                 'tree_method': 'gpu_exact',
                 'max_depth': 2,
                 'eval_metric': 'auc'}
        res = {}
@@ -134,7 +134,7 @@ class TestGPU(unittest.TestCase):
                            'objective': 'binary:logistic',
                            'eval_metric': 'auc'}
                ag_param2 = {'max_depth': max_depth,
-                             'updater': 'grow_gpu_hist',
+                             'tree_method': 'gpu_hist',
                             'eta': 1,
                             'silent': 1,
                             'n_gpus': 1,
@@ -142,7 +142,7 @@ class TestGPU(unittest.TestCase):
                                 'max_bin': max_bin,
                             'eval_metric': 'auc'}
                ag_param3 = {'max_depth': max_depth,
-                             'updater': 'grow_gpu_hist',
+                             'tree_method': 'gpu_hist',
                             'eta': 1,
                             'silent': 1,
                             'n_gpus': n_gpus,
@@ -177,7 +177,7 @@ class TestGPU(unittest.TestCase):
                dtest = xgb.DMatrix(X_test, y_test)

                param = {'objective': 'binary:logistic',
-                         'updater': 'grow_gpu_hist',
+                         'tree_method': 'gpu_hist',
                         'max_depth': max_depth,
                         'n_gpus': 1,
                         'max_bin': max_bin,
@@ -189,7 +189,7 @@ class TestGPU(unittest.TestCase):
                assert self.non_decreasing(res['train']['auc'])
                #assert self.non_decreasing(res['test']['auc'])
                param2 = {'objective': 'binary:logistic',
-                          'updater': 'grow_gpu_hist',
+                          'tree_method': 'gpu_hist',
                          'max_depth': max_depth,
                          'n_gpus': n_gpus,
                          'max_bin': max_bin,
@@ -211,7 +211,7 @@ class TestGPU(unittest.TestCase):
                dtrain2 = xgb.DMatrix(X2, label=y2)

                param = {'objective': 'binary:logistic',
-                         'updater': 'grow_gpu_hist',
+                         'tree_method': 'gpu_hist',
                         'max_depth': max_depth,
                         'n_gpus': n_gpus,
                         'max_bin': max_bin,
@@ -250,7 +250,7 @@ class TestGPU(unittest.TestCase):
                ######################################################################
                # fail-safe test for max_bin
                param = {'objective': 'binary:logistic',
-                         'updater': 'grow_gpu_hist',
+                         'tree_method': 'gpu_hist',
                         'max_depth': max_depth,
                         'n_gpus': n_gpus,
                         'eval_metric': 'auc',
@@ -263,7 +263,7 @@ class TestGPU(unittest.TestCase):
                ######################################################################
                # subsampling
                param = {'objective': 'binary:logistic',
-                         'updater': 'grow_gpu_hist',
+                         'tree_method': 'gpu_hist',
                         'max_depth': max_depth,
                         'n_gpus': n_gpus,
                         'eval_metric': 'auc',
@@ -279,7 +279,7 @@ class TestGPU(unittest.TestCase):
        ######################################################################
        # fail-safe test for max_bin=2
        param = {'objective': 'binary:logistic',
-                 'updater': 'grow_gpu_hist',
+                 'tree_method': 'gpu_hist',
                 'max_depth': 2,
                 'n_gpus': n_gpus,
                 'eval_metric': 'auc',