Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix GPU ID and prediction cache from pickle (#5086)

Browse Source 
* Hack for saving GPU ID.

* Declare prediction cache on GBTree.

* Add a simple test.

* Add `auto` option for GPU Predictor.
This commit is contained in:
Jiaming Yuan
2019-12-07 16:02:06 +08:00
committed by GitHub
parent 7ef5b78003
commit 608ebbe444
17 changed files with 362 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/cpp/gbm/test_gbtree.cc
View File

@@ -29,21 +29,17 @@ TEST(GBTree, SelectTreeMethod) {
ASSERT_EQ(tparam.updater_seq, "grow_colmaker,prune");
gbtree.Configure({{"tree_method", "hist"}, {"num_feature", n_feat}});
ASSERT_EQ(tparam.updater_seq, "grow_quantile_histmaker");
ASSERT_EQ(tparam.predictor, "cpu_predictor");
gbtree.Configure({{"booster", "dart"}, {"tree_method", "hist"},
{"num_feature", n_feat}});
ASSERT_EQ(tparam.updater_seq, "grow_quantile_histmaker");
ASSERT_EQ(tparam.predictor, "cpu_predictor");
#ifdef XGBOOST_USE_CUDA
generic_param.UpdateAllowUnknown(Args{{"gpu_id", "0"}});
gbtree.Configure({{"tree_method", "gpu_hist"}, {"num_feature", n_feat}});
ASSERT_EQ(tparam.updater_seq, "grow_gpu_hist");
ASSERT_EQ(tparam.predictor, "gpu_predictor");
gbtree.Configure({{"booster", "dart"}, {"tree_method", "gpu_hist"},
{"num_feature", n_feat}});
ASSERT_EQ(tparam.updater_seq, "grow_gpu_hist");
ASSERT_EQ(tparam.predictor, "gpu_predictor");
#endif
}

7
tests/cpp/predictor/test_cpu_predictor.cc
View File

@@ -9,8 +9,9 @@
namespace xgboost {
TEST(cpu_predictor, Test) {
auto lparam = CreateEmptyGenericParam(GPUIDX);
auto cache = std::make_shared<std::unordered_map<DMatrix*, PredictionCacheEntry>>();
std::unique_ptr<Predictor> cpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &lparam));
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &lparam, cache));
gbm::GBTreeModel model = CreateTestModel();
@@ -62,8 +63,10 @@ TEST(cpu_predictor, ExternalMemoryTest) {
std::string filename = tmpdir.path + "/big.libsvm";
std::unique_ptr<DMatrix> dmat = CreateSparsePageDMatrix(12, 64, filename);
auto lparam = CreateEmptyGenericParam(GPUIDX);
auto cache = std::make_shared<std::unordered_map<DMatrix*, PredictionCacheEntry>>();
std::unique_ptr<Predictor> cpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &lparam));
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &lparam, cache));
gbm::GBTreeModel model = CreateTestModel();

14
tests/cpp/predictor/test_gpu_predictor.cu
View File

@@ -36,14 +36,15 @@ namespace predictor {
TEST(gpu_predictor, Test) {
auto cpu_lparam = CreateEmptyGenericParam(-1);
auto gpu_lparam = CreateEmptyGenericParam(0);
auto cache = std::make_shared<std::unordered_map<DMatrix*, PredictionCacheEntry>>();
std::unique_ptr<Predictor> gpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("gpu_predictor", &gpu_lparam));
std::unique_ptr<Predictor>(Predictor::Create("gpu_predictor", &gpu_lparam, cache));
std::unique_ptr<Predictor> cpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &cpu_lparam));
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &cpu_lparam, cache));
gpu_predictor->Configure({}, {});
cpu_predictor->Configure({}, {});
gpu_predictor->Configure({});
cpu_predictor->Configure({});
for (size_t i = 1; i < 33; i *= 2) {
int n_row = i, n_col = i;
@@ -71,9 +72,10 @@ TEST(gpu_predictor, Test) {
TEST(gpu_predictor, ExternalMemoryTest) {
auto lparam = CreateEmptyGenericParam(0);
auto cache = std::make_shared<std::unordered_map<DMatrix*, PredictionCacheEntry>>();
std::unique_ptr<Predictor> gpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("gpu_predictor", &lparam));
gpu_predictor->Configure({}, {});
std::unique_ptr<Predictor>(Predictor::Create("gpu_predictor", &lparam, cache));
gpu_predictor->Configure({});
gbm::GBTreeModel model = CreateTestModel();
model.param.num_feature = 3;
const int n_classes = 3;

48
tests/python-gpu/test_gpu_training_continuation.py Normal file
View File

@@ -0,0 +1,48 @@
import unittest
import numpy as np
import xgboost as xgb
import json
rng = np.random.RandomState(1994)
class TestGPUTrainingContinuation(unittest.TestCase):
def test_training_continuation_binary(self):
kRows = 32
kCols = 16
X = np.random.randn(kRows, kCols)
y = np.random.randn(kRows)
dtrain = xgb.DMatrix(X, y)
params = {'tree_method': 'gpu_hist', 'max_depth': '2'}
bst_0 = xgb.train(params, dtrain, num_boost_round=4)
dump_0 = bst_0.get_dump(dump_format='json')
bst_1 = xgb.train(params, dtrain, num_boost_round=2)
bst_1 = xgb.train(params, dtrain, num_boost_round=2, xgb_model=bst_1)
dump_1 = bst_1.get_dump(dump_format='json')
def recursive_compare(obj_0, obj_1):
if isinstance(obj_0, float):
assert np.isclose(obj_0, obj_1)
elif isinstance(obj_0, str):
assert obj_0 == obj_1
elif isinstance(obj_0, int):
assert obj_0 == obj_1
elif isinstance(obj_0, dict):
keys_0 = list(obj_0.keys())
keys_1 = list(obj_1.keys())
values_0 = list(obj_0.values())
values_1 = list(obj_1.values())
for i in range(len(obj_0.items())):
assert keys_0[i] == keys_1[i]
if list(obj_0.keys())[i] != 'missing':
recursive_compare(values_0[i],
values_1[i])
else:
for i in range(len(obj_0)):
recursive_compare(obj_0[i], obj_1[i])
for i in range(len(dump_0)):
obj_0 = json.loads(dump_0[i])
obj_1 = json.loads(dump_1[i])
recursive_compare(obj_0, obj_1)

5
tests/python-gpu/test_gpu_with_dask.py
View File

@@ -45,9 +45,8 @@ class TestDistributedGPU(unittest.TestCase):
assert isinstance(out['booster'], dxgb.Booster)
assert len(out['history']['X']['rmse']) == 2
# FIXME(trivialfis): Re-enable this after #5003 is fixed
# predictions = dxgb.predict(client, out, dtrain).compute()
# assert isinstance(predictions, np.ndarray)
predictions = dxgb.predict(client, out, dtrain).compute()
assert isinstance(predictions, np.ndarray)
@pytest.mark.skipif(**tm.no_dask())
@pytest.mark.skipif(**tm.no_dask_cuda())