Skip to content

GitLab

Unverified Commit 191d99a7 authored by Yanhui Liang's avatar Yanhui Liang Committed by GitHub
Browse files

Make boosted_trees Garden-official (#4377) 

* Make boosted_trees Garden-official

* Fix nits
parent 1886043f
Show whitespace changes
Inline Side-by-side
Showing with 148 additions and 116 deletions
official/boosted_trees/README.md
View file @ 191d99a7
......@@ -39,7 +39,7 @@ Note that the model_dir is cleaned up before every time training starts.
Model parameters can be adjusted by flags, like `--n_trees`, `--max_depth`, `--learning_rate` and so on. Check out the code for details.
The final accuacy will be around 74% and loss will be around 0.516 over the eval set, when trained with the default parameters.
The final accuracy will be around 74% and loss will be around 0.516 over the eval set, when trained with the default parameters.
By default, the first 1 million examples among 11 millions are used for training, and the last 1 million examples are used for evaluation.
The training/evaluation data can be selected as index ranges by flags `--train_start`, `--train_count`, `--eval_start`, `--eval_count`, etc.
......
official/boosted_trees/data_download.py
View file @ 191d99a7
......@@ -12,28 +12,23 @@ from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import gzip
import os
import sys
import tempfile
# pylint: disable=g-bad-import-order
import numpy as np
import pandas as pd
from six.moves import urllib
from absl import app as absl_app
from absl import flags
import tensorflow as tf
URL_ROOT = 'https://archive.ics.uci.edu/ml/machine-learning-databases/00280'
INPUT_FILE = 'HIGGS.csv.gz'
NPZ_FILE = 'HIGGS.csv.gz.npz' # numpy compressed file to contain 'data' array.
from official.utils.flags import core as flags_core
def parse_args():
"""Parses arguments and returns a tuple (known_args, unparsed_args)."""
parser = argparse.ArgumentParser()
parser.add_argument(
'--data_dir', type=str, default='/tmp/higgs_data',
help='Directory to download higgs dataset and store training/eval data.')
return parser.parse_known_args()
URL_ROOT = "https://archive.ics.uci.edu/ml/machine-learning-databases/00280"
INPUT_FILE = "HIGGS.csv.gz"
NPZ_FILE = "HIGGS.csv.gz.npz" # numpy compressed file to contain "data" array.
def _download_higgs_data_and_save_npz(data_dir):
......@@ -41,30 +36,30 @@ def _download_higgs_data_and_save_npz(data_dir):
input_url = os.path.join(URL_ROOT, INPUT_FILE)
np_filename = os.path.join(data_dir, NPZ_FILE)
if tf.gfile.Exists(np_filename):
raise ValueError('data_dir already has the processed data file: {}'.format(
raise ValueError("data_dir already has the processed data file: {}".format(
np_filename))
if not tf.gfile.Exists(data_dir):
tf.gfile.MkDir(data_dir)
# 2.8 GB to download.
try:
print('Data downloading..')
tf.logging.info("Data downloading...")
temp_filename, _ = urllib.request.urlretrieve(input_url)
# Reading and parsing 11 million csv lines takes 2~3 minutes.
print('Data processing.. taking multiple minutes..')
tf.logging.info("Data processing... taking multiple minutes...")
with gzip.open(temp_filename, "rb") as csv_file:
data = pd.read_csv(
temp_filename,
csv_file,
dtype=np.float32,
names=['c%02d' % i for i in range(29)] # label + 28 features.
names=["c%02d" % i for i in range(29)] # label + 28 features.
).as_matrix()
finally:
os.remove(temp_filename)
tf.gfile.Remove(temp_filename)
# Writing to temporary location then copy to the data_dir (0.8 GB).
f = tempfile.NamedTemporaryFile()
np.savez_compressed(f, data=data)
tf.gfile.Copy(f.name, np_filename)
print('Data saved to: {}'.format(np_filename))
tf.logging.info("Data saved to: {}".format(np_filename))
def main(unused_argv):
......@@ -73,6 +68,16 @@ def main(unused_argv):
_download_higgs_data_and_save_npz(FLAGS.data_dir)
if __name__ == '__main__':
FLAGS, unparsed = parse_args()
tf.app.run(argv=[sys.argv[0]] + unparsed)
def define_data_download_flags():
"""Add flags specifying data download arguments."""
flags.DEFINE_string(
name="data_dir", default="/tmp/higgs_data",
help=flags_core.help_wrap(
"Directory to download higgs dataset and store training/eval data."))
if __name__ == "__main__":
tf.logging.set_verbosity(tf.logging.INFO)
define_data_download_flags()
FLAGS = flags.FLAGS
absl_app.run(main)
official/boosted_trees/train_higgs.py
View file @ 191d99a7
......@@ -29,64 +29,44 @@ from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import os
import sys
# pylint: disable=g-bad-import-order
import numpy as np
from absl import app as absl_app
from absl import flags
import numpy as np # pylint: disable=wrong-import-order
import tensorflow as tf # pylint: disable=wrong-import-order
import tensorflow as tf
# pylint: enable=g-bad-import-order
from official.utils.flags import core as flags_core
from official.utils.flags._conventions import help_wrap
from official.utils.logs import logger
NPZ_FILE = "HIGGS.csv.gz.npz" # numpy compressed file containing "data" array
NPZ_FILE = 'HIGGS.csv.gz.npz' # numpy compressed file containing 'data' array
def define_train_higgs_flags():
"""Add tree related flags as well as training/eval configuration."""
flags_core.define_base(stop_threshold=False, batch_size=False, num_gpu=False)
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_integer(
name='train_start', default=0,
help=help_wrap('Start index of train examples within the data.'))
flags.DEFINE_integer(
name='train_count', default=1000000,
help=help_wrap('Number of train examples within the data.'))
flags.DEFINE_integer(
name='eval_start', default=10000000,
help=help_wrap('Start index of eval examples within the data.'))
flags.DEFINE_integer(
name='eval_count', default=1000000,
help=help_wrap('Number of eval examples within the data.'))
flags.DEFINE_integer(
'n_trees', default=100, help=help_wrap('Number of trees to build.'))
flags.DEFINE_integer(
'max_depth', default=6, help=help_wrap('Maximum depths of each tree.'))
flags.DEFINE_float(
'learning_rate', default=0.1,
help=help_wrap('Maximum depths of each tree.'))
flags_core.set_defaults(data_dir='/tmp/higgs_data',
model_dir='/tmp/higgs_model')
def read_higgs_data(data_dir, train_start, train_count, eval_start, eval_count):
"""Reads higgs data from csv and returns train and eval data.
Args:
data_dir: A string, the directory of higgs dataset.
train_start: An integer, the start index of train examples within the data.
train_count: An integer, the number of train examples within the data.
eval_start: An integer, the start index of eval examples within the data.
eval_count: An integer, the number of eval examples within the data.
def read_higgs_data(data_dir, train_start, train_count, eval_start, eval_count):
"""Reads higgs data from csv and returns train and eval data."""
Returns:
Numpy array of train data and eval data.
"""
npz_filename = os.path.join(data_dir, NPZ_FILE)
try:
# gfile allows numpy to read data from network data sources as well.
with tf.gfile.Open(npz_filename, 'rb') as npz_file:
with tf.gfile.Open(npz_filename, "rb") as npz_file:
with np.load(npz_file) as npz:
data = npz['data']
data = npz["data"]
except Exception as e:
raise RuntimeError(
'Error loading data; use data_download.py to prepare the data:\n{}: {}'
"Error loading data; use data_download.py to prepare the data:\n{}: {}"
.format(type(e).__name__, e))
return (data[train_start:train_start+train_count],
data[eval_start:eval_start+eval_count])
......@@ -105,18 +85,18 @@ def make_inputs_from_np_arrays(features_np, label_np):
as a single tensor. Don't use batch.
Args:
features_np: a numpy ndarray (shape=[batch_size, num_features]) for
features_np: A numpy ndarray (shape=[batch_size, num_features]) for
float32 features.
label_np: a numpy ndarray (shape=[batch_size, 1]) for labels.
label_np: A numpy ndarray (shape=[batch_size, 1]) for labels.
Returns:
input_fn: a function returning a Dataset of feature dict and label.
feature_column: a list of tf.feature_column.BucketizedColumn.
input_fn: A function returning a Dataset of feature dict and label.
feature_column: A list of tf.feature_column.BucketizedColumn.
"""
num_features = features_np.shape[1]
features_np_list = np.split(features_np, num_features, axis=1)
# 1-based feature names.
feature_names = ['feature_%02d' % (i + 1) for i in range(num_features)]
feature_names = ["feature_%02d" % (i + 1) for i in range(num_features)]
# Create source feature_columns and bucketized_columns.
def get_bucket_boundaries(feature):
......@@ -155,15 +135,15 @@ def make_eval_inputs_from_np_arrays(features_np, label_np):
num_features = features_np.shape[1]
features_np_list = np.split(features_np, num_features, axis=1)
# 1-based feature names.
feature_names = ['feature_%02d' % (i + 1) for i in range(num_features)]
feature_names = ["feature_%02d" % (i + 1) for i in range(num_features)]
def input_fn():
features = {
feature_name: tf.constant(features_np_list[i])
for i, feature_name in enumerate(feature_names)
}
return tf.data.Dataset.zip(
(tf.data.Dataset.from_tensor_slices(features),
return tf.data.Dataset.zip((
tf.data.Dataset.from_tensor_slices(features),
tf.data.Dataset.from_tensor_slices(label_np),)).batch(1000)
return input_fn
......@@ -175,22 +155,37 @@ def train_boosted_trees(flags_obj):
Args:
flags_obj: An object containing parsed flag values.
"""
# Clean up the model directory if present.
if tf.gfile.Exists(flags_obj.model_dir):
tf.gfile.DeleteRecursively(flags_obj.model_dir)
print('## data loading..')
tf.logging.info("## Data loading...")
train_data, eval_data = read_higgs_data(
flags_obj.data_dir, flags_obj.train_start, flags_obj.train_count,
flags_obj.eval_start, flags_obj.eval_count)
print('## data loaded; train: {}{}, eval: {}{}'.format(
tf.logging.info("## Data loaded; train: {}{}, eval: {}{}".format(
train_data.dtype, train_data.shape, eval_data.dtype, eval_data.shape))
# data consists of one label column and 28 feature columns following.
# Data consists of one label column followed by 28 feature columns.
train_input_fn, feature_columns = make_inputs_from_np_arrays(
features_np=train_data[:, 1:], label_np=train_data[:, 0:1])
eval_input_fn = make_eval_inputs_from_np_arrays(
features_np=eval_data[:, 1:], label_np=eval_data[:, 0:1])
print('## features prepared. training starts..')
tf.logging.info("## Features prepared. Training starts...")
# Create benchmark logger to log info about the training and metric values
run_params = {
"train_start": flags_obj.train_start,
"train_count": flags_obj.train_count,
"eval_start": flags_obj.eval_start,
"eval_count": flags_obj.eval_count,
"n_trees": flags_obj.n_trees,
"max_depth": flags_obj.max_depth,
}
benchmark_logger = logger.config_benchmark_logger(flags_obj)
benchmark_logger.log_run_info(
model_name="boosted_trees",
dataset_name="higgs",
run_params=run_params)
# Though BoostedTreesClassifier is under tf.estimator, faster in-memory
# training is yet provided as a contrib library.
......@@ -203,7 +198,9 @@ def train_boosted_trees(flags_obj):
learning_rate=flags_obj.learning_rate)
# Evaluation.
eval_result = classifier.evaluate(eval_input_fn)
eval_results = classifier.evaluate(eval_input_fn)
# Benchmark the evaluation results
benchmark_logger.log_evaluation_result(eval_results)
# Exporting the savedmodel.
if flags_obj.export_dir is not None:
......@@ -216,7 +213,37 @@ def main(_):
train_boosted_trees(flags.FLAGS)
if __name__ == '__main__':
def define_train_higgs_flags():
"""Add tree related flags as well as training/eval configuration."""
flags_core.define_base(stop_threshold=False, batch_size=False, num_gpu=False)
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_integer(
name="train_start", default=0,
help=help_wrap("Start index of train examples within the data."))
flags.DEFINE_integer(
name="train_count", default=1000000,
help=help_wrap("Number of train examples within the data."))
flags.DEFINE_integer(
name="eval_start", default=10000000,
help=help_wrap("Start index of eval examples within the data."))
flags.DEFINE_integer(
name="eval_count", default=1000000,
help=help_wrap("Number of eval examples within the data."))
flags.DEFINE_integer(
"n_trees", default=100, help=help_wrap("Number of trees to build."))
flags.DEFINE_integer(
"max_depth", default=6, help=help_wrap("Maximum depths of each tree."))
flags.DEFINE_float(
"learning_rate", default=0.1,
help=help_wrap("The learning rate."))
flags_core.set_defaults(data_dir="/tmp/higgs_data",
model_dir="/tmp/higgs_model")
if __name__ == "__main__":
# Training progress and eval results are shown as logging.INFO; so enables it.
tf.logging.set_verbosity(tf.logging.INFO)
define_train_higgs_flags()
......
official/boosted_trees/train_higgs_test.py
View file @ 191d99a7
......@@ -12,7 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for boosted_tree."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
......@@ -22,16 +22,16 @@ import tempfile
import numpy as np
import pandas as pd
import tensorflow as tf
import tensorflow as tf # pylint: disable=g-bad-import-order
from official.utils.testing import integration
# pylint: disable=g-bad-import-order
from official.boosted_trees import train_higgs
from official.utils.testing import integration
tf.logging.set_verbosity(tf.logging.ERROR)
TEST_CSV = os.path.join(os.path.dirname(__file__), "train_higgs_test.csv")
tf.logging.set_verbosity(tf.logging.ERROR)
TEST_CSV = os.path.join(os.path.dirname(__file__), 'train_higgs_test.csv')
class BaseTest(tf.test.TestCase):
"""Tests for Wide Deep model."""
......@@ -45,7 +45,7 @@ class BaseTest(tf.test.TestCase):
# Create temporary CSV file
self.data_dir = self.get_temp_dir()
data = pd.read_csv(
TEST_CSV, dtype=np.float32, names=['c%02d' % i for i in range(29)]
TEST_CSV, dtype=np.float32, names=["c%02d" % i for i in range(29)]
).as_matrix()
self.input_npz = os.path.join(self.data_dir, train_higgs.NPZ_FILE)
# numpy.savez doesn't take gfile.Gfile, so need to write down and copy.
......@@ -56,9 +56,9 @@ class BaseTest(tf.test.TestCase):
def test_read_higgs_data(self):
"""Tests read_higgs_data() function."""
# Error when a wrong data_dir is given.
with self.assertRaisesRegexp(RuntimeError, 'Error loading data.*'):
with self.assertRaisesRegexp(RuntimeError, "Error loading data.*"):
train_data, eval_data = train_higgs.read_higgs_data(
self.data_dir + 'non-existing-path',
self.data_dir + "non-existing-path",
train_start=0, train_count=15, eval_start=15, eval_count=5)
# Loading fine with the correct data_dir.
......@@ -80,13 +80,13 @@ class BaseTest(tf.test.TestCase):
self.assertEqual(28, len(feature_columns))
bucketized_column_type = type(
tf.feature_column.bucketized_column(
tf.feature_column.numeric_column('feature_01'),
tf.feature_column.numeric_column("feature_01"),
boundaries=[0, 1, 2])) # dummy boundaries.
for feature_column in feature_columns:
self.assertIsInstance(feature_column, bucketized_column_type)
# At least 2 boundaries.
self.assertGreaterEqual(len(feature_column.boundaries), 2)
feature_names = ['feature_%02d' % (i+1) for i in range(28)]
feature_names = ["feature_%02d" % (i+1) for i in range(28)]
# Tests that the source column names of the bucketized columns match.
self.assertAllEqual(feature_names,
[col.source_column.name for col in feature_columns])
......@@ -113,39 +113,39 @@ class BaseTest(tf.test.TestCase):
def test_end_to_end(self):
"""Tests end-to-end running."""
model_dir = os.path.join(self.get_temp_dir(), 'model')
model_dir = os.path.join(self.get_temp_dir(), "model")
integration.run_synthetic(
main=train_higgs.main, tmp_root=self.get_temp_dir(), extra_flags=[
'--data_dir', self.data_dir,
'--model_dir', model_dir,
'--n_trees', '5',
'--train_start', '0',
'--train_count', '12',
'--eval_start', '12',
'--eval_count', '8',
"--data_dir", self.data_dir,
"--model_dir", model_dir,
"--n_trees", "5",
"--train_start", "0",
"--train_count", "12",
"--eval_start", "12",
"--eval_count", "8",
],
synth=False, max_train=None)
self.assertTrue(tf.gfile.Exists(os.path.join(model_dir, 'checkpoint')))
self.assertTrue(tf.gfile.Exists(os.path.join(model_dir, "checkpoint")))
def test_end_to_end_with_export(self):
"""Tests end-to-end running."""
model_dir = os.path.join(self.get_temp_dir(), 'model')
export_dir = os.path.join(self.get_temp_dir(), 'export')
model_dir = os.path.join(self.get_temp_dir(), "model")
export_dir = os.path.join(self.get_temp_dir(), "export")
integration.run_synthetic(
main=train_higgs.main, tmp_root=self.get_temp_dir(), extra_flags=[
'--data_dir', self.data_dir,
'--model_dir', model_dir,
'--export_dir', export_dir,
'--n_trees', '5',
'--train_start', '0',
'--train_count', '12',
'--eval_start', '12',
'--eval_count', '8',
"--data_dir", self.data_dir,
"--model_dir", model_dir,
"--export_dir", export_dir,
"--n_trees", "5",
"--train_start", "0",
"--train_count", "12",
"--eval_start", "12",
"--eval_count", "8",
],
synth=False, max_train=None)
self.assertTrue(tf.gfile.Exists(os.path.join(model_dir, 'checkpoint')))
self.assertTrue(tf.gfile.Exists(os.path.join(model_dir, "checkpoint")))
self.assertTrue(tf.gfile.Exists(os.path.join(export_dir)))
if __name__ == '__main__':
if __name__ == "__main__":
tf.test.main()
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment