Bug fix and refine cloud training for DS2.

Summary:
1. Add missing is_local argument (when set False, use pserver).
2. Add exception thrown if cp failed.
3. Add cloud mkdir if not cloud path for uploading does not exist.
4. Fix a bug using common path ./local_manifest for all nodes. (convert to /local_manifest)
5. Refine coding style.

cloud/_init_paths.py

@@ -0,0 +1,17 @@
+"""Set up paths for DS2"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os.path
+import sys
+
+
+def add_path(path):
+    if path not in sys.path:
+        sys.path.insert(0, path)
+
+
+this_dir = os.path.dirname(__file__)
+proj_path = os.path.join(this_dir, '..')
+add_path(proj_path)

cloud/pcloud_submit.sh

@@ -1,17 +1,22 @@
 # Configure input data set in local filesystem
-TRAIN_MANIFEST="/home/work/demo/ds2/pcloud/models/deep_speech_2/datasets/manifest.dev"
-TEST_MANIFEST="/home/work/demo/ds2/pcloud/models/deep_speech_2/datasets/manifest.dev"
-VOCAB_FILE="/home/work/demo/ds2/pcloud/models/deep_speech_2/datasets/vocab/eng_vocab.txt"
-MEAN_STD_FILE="/home/work/demo/ds2/pcloud/models/deep_speech_2/mean_std.npz"
-
+TRAIN_MANIFEST="../datasets/manifest.dev"
+DEV_MANIFEST="../datasets/manifest.dev"
+VOCAB_FILE="../datasets/vocab/eng_vocab.txt"
+MEAN_STD_FILE="../mean_std.npz"
 # Configure output path in PaddleCloud filesystem
-CLOUD_DATA_DIR="/pfs/dlnel/home/demo/deepspeech2/data"
-CLOUD_MODEL_DIR="/pfs/dlnel/home/demo/deepspeech2/model"
+CLOUD_DATA_DIR="/pfs/dlnel/home/sunxinghai@baidu.com/deepspeech2/data"
+CLOUD_MODEL_DIR="/pfs/dlnel/home/sunxinghai@baidu.com/deepspeech2/model"
+# Configure cloud resources
+NUM_CPU=12
+NUM_GPU=4
+NUM_NODE=2
+MEMORY="10Gi"
+IS_LOCAL="False"
 
 # Pack and upload local data to PaddleCloud filesystem
 python upload_data.py \
 --train_manifest_path=${TRAIN_MANIFEST} \
---test_manifest_path=${TEST_MANIFEST} \
+--dev_manifest_path=${DEV_MANIFEST} \
 --vocab_file=${VOCAB_FILE} \
 --mean_std_file=${MEAN_STD_FILE} \
 --cloud_data_path=${CLOUD_DATA_DIR}
@@ -21,23 +26,23 @@ then
     exit 1
 fi
 
-JOB_NAME=deepspeech`date +%Y%m%d%H%M%S`
+# Submit job to PaddleCloud
+JOB_NAME=deepspeech-`date +%Y%m%d%H%M%S`
 DS2_PATH=${PWD%/*}
 cp -f  pcloud_train.sh ${DS2_PATH}
 
-# Configure computation resource and submit job to PaddleCloud
 paddlecloud submit \
 -image bootstrapper:5000/wanghaoshuang/pcloud_ds2:latest \
 -jobname ${JOB_NAME} \
--cpu 4 \
--gpu 4 \
--memory 10Gi \
--parallelism 2 \
+-cpu ${NUM_CPU} \
+-gpu ${NUM_GPU} \
+-memory ${MEMORY} \
+-parallelism ${NUM_NODE} \
 -pscpu 1 \
 -pservers 1 \
--psmemory 10Gi \
+-psmemory ${MEMORY} \
 -passes 1 \
--entry "sh pcloud_train.sh ${CLOUD_DATA_DIR} ${CLOUD_MODEL_DIR}" \
+-entry "sh pcloud_train.sh ${CLOUD_DATA_DIR} ${CLOUD_MODEL_DIR} ${NUM_CPU} ${NUM_GPU} ${IS_LOCAL}" \
 ${DS2_PATH}
 
 rm ${DS2_PATH}/pcloud_train.sh

cloud/pcloud_train.sh

@@ -1,28 +1,36 @@
 DATA_PATH=$1
 MODEL_PATH=$2
+NUM_CPU=$3
+NUM_GPU=$4
+IS_LOCAL=$5
+
 TRAIN_MANI=${DATA_PATH}/cloud.train.manifest
-DEV_MANI=${DATA_PATH}/cloud.test.manifest
+DEV_MANI=${DATA_PATH}/cloud.dev.manifest
 TRAIN_TAR=${DATA_PATH}/cloud.train.tar
-DEV_TAR=${DATA_PATH}/cloud.test.tar
+DEV_TAR=${DATA_PATH}/cloud.dev.tar
 VOCAB_PATH=${DATA_PATH}/vocab.txt
 MEAN_STD_FILE=${DATA_PATH}/mean_std.npz
 
 # split train data for each pcloud node
 python ./cloud/split_data.py \
---in_manifest_path=$TRAIN_MANI \
---data_tar_path=$TRAIN_TAR \
---out_manifest_path='./local.train.manifest'
+--in_manifest_path=${TRAIN_MANI} \
+--data_tar_path=${TRAIN_TAR} \
+--out_manifest_path='/local.train.manifest'
 
 # split dev data for each pcloud node
 python ./cloud/split_data.py \
---in_manifest_path=$DEV_MANI \
---data_tar_path=$DEV_TAR \
---out_manifest_path='./local.test.manifest'
+--in_manifest_path=${DEV_MANI} \
+--data_tar_path=${DEV_TAR} \
+--out_manifest_path='/local.dev.manifest'
 
+# run train
 python train.py \
 --use_gpu=1 \
---mean_std_filepath=$MEAN_STD_FILE \
---train_manifest_path='./local.train.manifest' \
---dev_manifest_path='./local.test.manifest' \
---vocab_filepath=$VOCAB_PATH \
+--trainer_count=${NUM_GPU} \
+--num_threads_data=${NUM_CPU} \
+--is_local=${IS_LOCAL} \
+--mean_std_filepath=${MEAN_STD_FILE} \
+--train_manifest_path='/local.train.manifest' \
+--dev_manifest_path='/local.dev.manifest' \
+--vocab_filepath=${VOCAB_PATH} \
 --output_model_dir=${MODEL_PATH}

cloud/split_data.py

@@ -6,6 +6,7 @@ This script should be called in paddle cloud.
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
+
 import os
 import json
 import argparse

cloud/upload_data.py

@@ -1,30 +1,31 @@
-"""This tool is used for preparing data for DeepSpeech2 trainning on paddle cloud.
+"""This script is used for preparing data for DeepSpeech2 trainning on paddle
+cloud.
 
 Steps:
 1. Read original manifest and get the local path of sound files.
 2. Tar all local sound files into one tar file.
 3. Modify original manifest to remove the local path information.
 
-Finally, we will get a tar file and a manifest with sound file name, duration
-and text.
+Finally, we will get a tar file and a new manifest.
 """
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
+
 import json
 import os
 import tarfile
 import sys
 import argparse
 import shutil
-sys.path.append('../')
-from data_utils.utils import read_manifest
 from subprocess import call
+import _init_paths
+from data_utils.utils import read_manifest
 
 TRAIN_TAR = "cloud.train.tar"
 TRAIN_MANIFEST = "cloud.train.manifest"
-TEST_TAR = "cloud.test.tar"
-TEST_MANIFEST = "cloud.test.manifest"
+DEV_TAR = "cloud.dev.tar"
+DEV_MANIFEST = "cloud.dev.manifest"
 VOCAB_FILE = "vocab.txt"
 MEAN_STD_FILE = "mean_std.npz"
 
@@ -33,29 +34,29 @@ parser.add_argument(
     "--train_manifest_path",
     default="../datasets/manifest.train",
     type=str,
-    help="Manifest file of train data. (default: %(default)s)")
+    help="Manifest file path for train data. (default: %(default)s)")
 parser.add_argument(
-    "--test_manifest_path",
-    default="../datasets/manifest.test",
+    "--dev_manifest_path",
+    default="../datasets/manifest.dev",
     type=str,
-    help="Manifest file of test data. (default: %(default)s)")
+    help="Manifest file path for validation data. (default: %(default)s)")
 parser.add_argument(
     "--vocab_file",
     default="../datasets/vocab/eng_vocab.txt",
     type=str,
-    help="Vocab file to be uploaded to paddlecloud. (default: %(default)s)")
+    help="Vocabulary file to be uploaded to paddlecloud. "
+    "(default: %(default)s)")
 parser.add_argument(
     "--mean_std_file",
     default="../mean_std.npz",
     type=str,
-    help="mean_std file to be uploaded to paddlecloud. (default: %(default)s)")
+    help="Normalizer's statistics (mean and stddev) file to be uploaded to "
+    "paddlecloud. (default: %(default)s)")
 parser.add_argument(
     "--cloud_data_path",
     required=True,
     type=str,
     help="Destination path on paddlecloud. (default: %(default)s)")
-args = parser.parse_args()
-
 parser.add_argument(
     "--local_tmp_path",
     default="./tmp/",
@@ -65,9 +66,9 @@ args = parser.parse_args()
 
 
 def pack_data(manifest_path, out_tar_path, out_manifest_path):
-    '''1. According to the manifest, tar sound files into out_tar_path
-    2. Generate a new manifest for output tar file
-    '''
+    """1. According to the manifest, tar sound files into out_tar_path.
+    2. Generate a new manifest for output tar file.
+    """
     out_tar = tarfile.open(out_tar_path, 'w')
     manifest = read_manifest(manifest_path)
     results = []
@@ -83,11 +84,19 @@ def pack_data(manifest_path, out_tar_path, out_manifest_path):
     out_tar.close()
 
 
+def pcloud_mkdir(dir):
+    """Make directory in PaddleCloud filesystem.
+    """
+    if call(['paddlecloud', 'mkdir', dir]) != 0:
+        raise IOError("PaddleCloud mkdir failed: %s." % dir)
+
+
 def pcloud_cp(src, dst):
-    """Copy src from local filesytem to dst in PaddleCloud filesystem.
+    """Copy src from local filesytem to dst in PaddleCloud filesystem,
+    or downlowd src from PaddleCloud filesystem to dst in local filesystem.
     """
-    ret = call(['paddlecloud', 'cp', src, dst])
-    return ret
+    if call(['paddlecloud', 'cp', src, dst]) != 0:
+        raise IOError("PaddleCloud cp failed: from [%s] to [%s]." % (src, dst))
 
 
 def pcloud_exist(path):
@@ -100,48 +109,34 @@ def pcloud_exist(path):
 if __name__ == '__main__':
     cloud_train_manifest = os.path.join(args.cloud_data_path, TRAIN_MANIFEST)
     cloud_train_tar = os.path.join(args.cloud_data_path, TRAIN_TAR)
-    cloud_test_manifest = os.path.join(args.cloud_data_path, TEST_MANIFEST)
-    cloud_test_tar = os.path.join(args.cloud_data_path, TEST_TAR)
+    cloud_dev_manifest = os.path.join(args.cloud_data_path, DEV_MANIFEST)
+    cloud_dev_tar = os.path.join(args.cloud_data_path, DEV_TAR)
     cloud_vocab_file = os.path.join(args.cloud_data_path, VOCAB_FILE)
     cloud_mean_file = os.path.join(args.cloud_data_path, MEAN_STD_FILE)
 
     local_train_manifest = os.path.join(args.local_tmp_path, TRAIN_MANIFEST)
     local_train_tar = os.path.join(args.local_tmp_path, TRAIN_TAR)
-    local_test_manifest = os.path.join(args.local_tmp_path, TEST_MANIFEST)
-    local_test_tar = os.path.join(args.local_tmp_path, TEST_TAR)
+    local_dev_manifest = os.path.join(args.local_tmp_path, DEV_MANIFEST)
+    local_dev_tar = os.path.join(args.local_tmp_path, DEV_TAR)
 
+    # prepare local and cloud dir
     if os.path.exists(args.local_tmp_path):
         shutil.rmtree(args.local_tmp_path)
     os.makedirs(args.local_tmp_path)
+    pcloud_mkdir(args.cloud_data_path)
 
-    # train data
-    if args.train_manifest_path != "":
-        ret = pcloud_exist(cloud_train_manifest)
-        if ret != 0:
-            pack_data(args.train_manifest_path, local_train_tar,
-                      local_train_manifest)
+    # pack and upload train data
+    pack_data(args.train_manifest_path, local_train_tar, local_train_manifest)
     pcloud_cp(local_train_manifest, cloud_train_manifest)
     pcloud_cp(local_train_tar, cloud_train_tar)
 
-    # test data
-    if args.test_manifest_path != "":
-        ret = pcloud_exist(cloud_test_manifest)
-        if ret != 0:
-            pack_data(args.test_manifest_path, local_test_tar,
-                      local_test_manifest)
-            pcloud_cp(local_test_manifest, cloud_test_manifest)
-            pcloud_cp(local_test_tar, cloud_test_tar)
-
-    # vocab file
-    if args.vocab_file != "":
-        ret = pcloud_exist(cloud_vocab_file)
-        if ret != 0:
-            pcloud_cp(args.vocab_file, cloud_vocab_file)
+    # pack and upload validation data
+    pack_data(args.dev_manifest_path, local_dev_tar, local_dev_manifest)
+    pcloud_cp(local_dev_manifest, cloud_dev_manifest)
+    pcloud_cp(local_dev_tar, cloud_dev_tar)
 
-    # mean_std file
-    if args.mean_std_file != "":
-        ret = pcloud_exist(cloud_mean_file)
-        if ret != 0:
+    # upload vocab file and mean_std file
+    pcloud_cp(args.vocab_file, cloud_vocab_file)
     pcloud_cp(args.mean_std_file, cloud_mean_file)
 
     shutil.rmtree(args.local_tmp_path)

train.py

@@ -127,6 +127,12 @@ parser.add_argument(
     type=str,
     help="Augmentation configuration in json-format. "
     "(default: %(default)s)")
+parser.add_argument(
+    "--is_local",
+    default=True,
+    type=distutils.util.strtobool,
+    help="Set to false if running with pserver in paddlecloud. "
+    "(default: %(default)s)")
 args = parser.parse_args()
 
 
@@ -178,7 +184,10 @@ def train():
 
 def main():
     utils.print_arguments(args)
-    paddle.init(use_gpu=args.use_gpu, trainer_count=args.trainer_count)
+    paddle.init(
+        use_gpu=args.use_gpu,
+        trainer_count=args.trainer_count,
+        is_local=args.is_local)
     train()