def test_two_phase_train_c2(self):
"""test single train with two phase in runner config.
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["phases"] = ['phase1', 'phase2']
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 4, self.err_msg)
def test_single_infer_gpu(self):
"""test single infer with gpu."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = 'runner2'
self.yaml_content["runner"][2]["device"] = 'gpu'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.run_time_re, 1, self.err_msg)
def test_multiview_simnet_gpu(self):
"""test match.multiview_simnet with gpu."""
self.yaml_config_name = "models/match/multiview-simnet/config.yaml"
sed_gpu_cmd = "sed -i 's/device:.*cpu/device: gpu/g' {}".format(
self.yaml_config_name)
utils.cmd_shell(sed_gpu_cmd)
self.run_yaml(generate=False, cuda_devices="0")
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_debug_open_gpu(self):
"""test debug open with gpu."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["device"] = 'gpu'
self.yaml_content["debug"] = 'True'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 2, self.err_msg)
def test_optimizer_sgd_gpu(self):
"""test optimizer sgd with gpu."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["device"] = 'gpu'
self.yaml_content["hyper_parameters"]['optimizer']['class'] = 'SGD'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 2, self.err_msg)
def test_single_infer(self):
"""test single infer."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = 'runner2'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, 'Infer.+done', 1, self.err_msg)
def test_increment_train(self):
"""test increment train."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = 'runner1'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 2, self.err_msg)
built_in.regex_match_equal(self.out, '\nmode\s+(\S+)\s+\n', 'runner1', self.err_msg)
def test_optimizer_sgd_reg(self):
"""test optimizer sgd with reg."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["hyper_parameters"]['optimizer']['class'] = 'SGD'
self.yaml_content["hyper_parameters"]['reg'] = 0.1
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, 'epoch.+done', 2, self.err_msg)
def test_tagspace_gpu(self):
"""test contentunderstanding.tagspace with gpu."""
self.yaml_config_name = "models/contentunderstanding/tagspace/config.yaml"
sed_gpu_cmd = "sed -i 's/device:.*cpu/device: gpu/g' {}".format(
self.yaml_config_name)
utils.cmd_shell(sed_gpu_cmd)
self.run_yaml(generate=False, cuda_devices="0")
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_single_infer_in_epochs_dir_c2(self):
"""test single infer base on one epoch with new config.
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = 'runner2'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.run_time_re, 1, self.err_msg)
def test_xdeepfm_gpu(self):
"""test rank.xdeepfm with gpu."""
self.yaml_config_name = "models/rank/xdeepfm/config.yaml"
sed_gpu_cmd = "sed -i 's/device:.*cpu/device: gpu/g' {}".format(
self.yaml_config_name)
utils.cmd_shell(sed_gpu_cmd)
self.run_yaml(generate=False, cuda_devices="0")
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_fibinet(self):
"""test rank.fibinet"""
self.yaml_config_name = "models/rank/fibinet/config.yaml"
verify_epochs = "sed -i 's/epochs:.*4/epochs: 1/g' {}".format(
self.yaml_config_name)
utils.cmd_shell(verify_epochs)
self.run_yaml(generate=False)
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_mode_str_ps_local_cluster_1p_1t_async_c2(self):
"""test_mode_str_ps_local_cluster_1p_1t_c2."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["mode"] = "runner0"
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains('logs/server.0', 'Traceback', self.err_msg)
built_in.contains('logs/worker.0', 'AsyncCommunicator Initialized', self.err_msg)
def test_single_infer_in_base_dir_c2(self):
"""test single infer base on save dir with new config.
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = 'runner2'
self.yaml_content['runner'][2]['init_model_path'] = 'increment_dnn'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_mode_list_ps_local_cluster_c2(self):
"""test mode list has one element and the runner is local cluster.
"""
utils.cmd_shell("rm -rf logs")
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_single_infer(self):
"""test single infer."""
# run basic to save models
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.run_yaml()
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = 'runner2'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_increment_train_c2(self):
"""test increment train.
both runners are single & train.
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['mode'] = ['runner0', 'runner1']
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 4, self.err_msg)
def test_AutoInt_gpu(self):
"""test rank.fibinet"""
self.yaml_config_name = "models/rank/AutoInt/config.yaml"
verify_epochs = "sed -i 's/epochs:.*2/epochs: 1/g' {}".format(self.yaml_config_name)
sed_gpu_cmd = "sed -i 's/device:.*cpu/device: gpu/g' {}".format(self.yaml_config_name)
utils.cmd_shell(sed_gpu_cmd)
utils.cmd_shell(verify_epochs)
self.run_yaml(generate=False, cuda_devices="0")
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
def test_mode_str_ps_local_cluster_1p_2t_sync_c2(self):
"""test_mode_str_ps_local_cluster_1p_2t_sync_c2."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["mode"] = "runner0"
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.yaml_content["runner"][0]["distribute_strategy"] = "sync"
self.yaml_content["runner"][0]["worker_num"] = 2
self.yaml_content["runner"][0]["server_num"] = 1
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains('logs/server.0', 'Traceback', self.err_msg)
def test_mode_list_ps_local_cluster_and_increment_c2(self):
"""test mode list has two elements and both runner are local cluster.
one is train, the other is increment training.
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.yaml_content["runner"][1]["class"] = "local_cluster_train"
self.yaml_content["mode"] = ["runner0", "runner1"]
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len('logs/worker.0', self.epoch_re, 2, self.err_msg)
def test_two_phase_train(self):
"""test two phase train"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['phase'].append({
'name': 'phase2',
'model': '{workspace}/model.py', # user-defined model
'dataset_name': 'dataset_infer', # select dataset by name
'thread_num': 1
})
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, 'epoch.+done', 4, self.err_msg)
def test_mode_list_single_selected_gpus_1card_c2(self):
"""test selected gpus 1card, it will run with single mode."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["device"] = 'gpu'
self.yaml_content["runner"][0]["selected_gpus"] = "0"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 2, self.err_msg)
built_in.regex_match_equal(self.out,
'\ntrain.trainer.engine\s+(\S+)\s+\n',
"single",
self.err_msg)
def test_QueueDataset_train_c2(self):
"""test QueueDataset in train."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["dataset"][0]["type"] = "QueueDataset"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 2, self.err_msg)
# NOTE windows和mac直接会强行切换到dataloader
if utils.get_platform() != "LINUX" or not six.PY2:
check_type = "DataLoader"
else:
check_type = "QueueDataset"
def test_mode_str_ps_local_cluster_2p_2t_async_c2(self):
"""test_mode_str_ps_local_cluster_2p_2t_async_c2."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["mode"] = "runner0"
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.yaml_content["runner"][0]["worker_num"] = 2
self.yaml_content["runner"][0]["server_num"] = 2
self.yaml_content["dataset"][0]["data_path"] = "criteo_data"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains('logs/server.0', 'Traceback', self.err_msg)
built_in.not_contains('logs/server.1', 'Traceback', self.err_msg)
built_in.regex_match_len('logs/worker.0', self.epoch_re, 2, self.err_msg)
built_in.regex_match_len('logs/worker.1', self.epoch_re, 2, self.err_msg)
def test_optimizer_lr_le(self):
"""test optimizer lr"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["hyper_parameters"]['optimizer']['class'] = 'SGD'
self.yaml_content["hyper_parameters"]['optimizer']['learning_rate'] = 2e-2
self.yaml_content["hyper_parameters"]['reg'] = 0.1
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.epoch_re, 2, self.err_msg)
built_in.regex_match_equal(self.out,
'\nhyper_parameters.optimizer.learning_rate\s+(\S+)\s+\n',
'0.02',
self.err_msg)
def test_mode_list_collective_selected_gpus_2f_2cards_c2(self):
"""test selected gpus 2card with two files and set fleet mode = collective,
it will change ps to collective and run with local_cluster_train mode
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["device"] = 'gpu'
self.yaml_content["runner"][0]["selected_gpus"] = "0,1"
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.yaml_content["runner"][0]["fleet_mode"] = "collective"
self.yaml_content["dataset"][0]["data_path"] = "criteo_data"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len('logs/worker.0', self.epoch_re, 2, self.err_msg)
built_in.path_not_exist('logs/server.0', self.err_msg)
def test_thread_num(self):
"""test thread num."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content['phase'].append({
'name': 'phase2',
'model': '{workspace}/model.py', # user-defined model
'dataset_name': 'dataset_infer', # select dataset by name
'thread_num': 2
})
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len(self.out, self.run_time_re, 4, self.err_msg)
built_in.regex_match_equal(self.out,
'\nphase.phase2.thread_num\s+(\S+)\s+\n', 2,
self.err_msg)
def test_user_define_all_normal(self):
"""test normal yaml construct by MultiTaskMMOE base."""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["instance_class_path"] = 'paddle_rec_user_define.py'
self.yaml_content["runner"][0]["network_class_path"] = 'paddle_rec_user_define.py'
self.yaml_content["runner"][0]["startup_class_path"] = 'paddle_rec_user_define.py'
self.yaml_content["runner"][0]["runner_class_path"] = 'paddle_rec_user_define.py'
self.yaml_content["runner"][0]["terminal_class_path"] = 'paddle_rec_user_define.py'
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.contains(self.out, 'User Define SingleInstance', self.err_msg)
built_in.contains(self.out, 'User Define SingleNetwork', self.err_msg)
built_in.contains(self.out, 'User Define SingleStartup', self.err_msg)
built_in.contains(self.out, 'User Define SingleRunner', self.err_msg)
built_in.contains(self.out, 'User Define SingleTerminal', self.err_msg)
def test_mode_list_collective_selected_gpus_2f_4cards_c2(self):
"""
test_collective_selected_gpus_2f_2cards.
程序运行GPU卡号，会依据文件个数, gpu_nums, worker_num 来判断起多少个trainer.
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["device"] = 'gpu'
self.yaml_content["runner"][0]["selected_gpus"] = "0,1"
self.yaml_content["runner"][0]["class"] = "local_cluster_train"
self.yaml_content["runner"][0]["fleet_mode"] = "collective"
self.yaml_content["dataset"][0]["data_path"] = "criteo_data"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.regex_match_len('logs/worker.1', self.epoch_re, 2, self.err_msg)
built_in.path_not_exist('logs/worker.2', self.err_msg)
def test_mode_list_ps_selected_gpus_2f_2card_c2(self):
"""test selected gpus 2card with two files and not set fleet mode,
it will change ps to collective and run with local_cluster_train mode
"""
self.yaml_config_name = sys._getframe().f_code.co_name + '.yaml'
self.yaml_content["runner"][0]["device"] = 'gpu'
self.yaml_content["runner"][0]["selected_gpus"] = "0,1"
self.yaml_content["dataset"][0]["data_path"] = "criteo_data"
self.run_yaml()
built_in.equals(self.pro.returncode, 0, self.err_msg)
built_in.not_contains(self.err, 'Traceback', self.err_msg)
built_in.path_not_exist('logs/server.0', self.err_msg)
built_in.regex_match_equal(self.out,
'\ntrain.trainer.engine\s+(\S+)\s+\n',
"local_cluster",
self.err_msg)
built_in.regex_match_len('logs/worker.1', self.auc_re, 6, self.err_msg)
