mindspore/model_zoo/official/recommend/naml

• Contents
• NAML Description
• Dataset
• Environment Requirements
• Script description
• • Script and sample code
  • Training process
  • • Usage
  • Model Export
• Model Description
• • Performance
  • • Evaluation Performance
    • Inference Performance
    • Inference on Ascend310 Performance
• Description of Random Situation
• ModelZoo Homepage

Contents

• NAML Description
• Dataset
• Environment Requirements
• Script Description
  • Script and Sample Code
    • Training Process
    • Model Export
• Model Description
  • Performance
    • Evaluation Performance
    • Inference Performance
• Description of Random Situation
• ModelZoo Homepage

NAML Description

NAML is a multi-view news recommendation approach. The core of NAML is a news encoder and a user encoder. The newsencoder is composed of a title encoder, a abstract encoder, a category encoder and a subcategory encoder. In the user encoder, we learn representations of users from their browsed news. Besides, we apply additive attention to learn more informative news and user representations by selecting important words and news.

Paper Chuhan Wu, Fangzhao Wu, Mingxiao An, Jianqiang Huang, Yongfeng Huang and Xing Xie: Neural News Recommendation with Attentive Multi-View Learning, IJCAI 2019

Dataset

Dataset used: MIND

MIND contains about 160k English news articles and more than 15 million impression logs generated by 1 million users.

You can download the dataset and put the directory in structure as follows:

└─MINDlarge
  ├─MINDlarge_train
  ├─MINDlarge_dev
  └─MINDlarge_utils

Environment Requirements

• Hardware（Ascend/GPU）
  • Prepare hardware environment with Ascend, GPU processor.
• Framework
  • MindSpore
• For more information, please check the resources below：
  • MindSpore Tutorials
  • MindSpore Python API

Script description

Script and sample code

├── naml
  ├── README.md                    # descriptions about NAML
  ├── scripts
  │   ├──run_train.sh              # shell script for training
  │   ├──run_eval.sh               # shell script for evaluation
  │   ├──run_infer_310.sh          # shell script for 310 inference
  ├── src
  │   ├──option.py                 # parse args
  │   ├──callback.py               # callback file
  │   ├──dataset.py                # creating dataset
  │   ├──naml.py                   # NAML architecture
  │   ├──config.py                 # config file
  │   ├──utils.py                  # utils to load ckpt_file for fine tune or incremental learn
  ├──ascend310_infer               #application for 310 inference
  ├── train.py                     # training script
  ├── eval.py                      # evaluation script
  ├── export.py                    # export mindir script
  └──postprogress.py               # post process for 310 inference

Training process

Usage

You can start training using python or shell scripts. The usage of shell scripts as follows:

# train standalone
bash run_train.sh [PLATFORM] [DEVICE_ID] [DATASET] [DATASET_PATH]
# train distribute
bash run_distribute_train.sh [PLATFORM] [DEVICE_NUM] [DATASET] [DATASET_PATH] [RANK_TABLE_FILE]
# evaluation
bash run_eval.sh [PLATFORM] [DEVICE_ID] [DATASET] [DATASET_PATH] [CHECKPOINT_PATH]

• PLATFORM should be Ascend.
• DEVICE_ID is the device id you want to run the network.
• DATASET MIND dataset, support large, small and demo.
• DATASET_PATH is the dataset path, the structure as Dataset.
• CHECKPOINT_PATH is a pre-trained checkpoint path.
• RANK_TABLE_FILE is HCCL configuration file when running on Ascend.

Model Export

python export.py --platform [PLATFORM] --checkpoint_path [CHECKPOINT_PATH] --file_format [EXPORT_FORMAT] --batch_size [BATCH_SIZE]

• EXPORT_FORMAT should be in ["AIR", "MINDIR"]

Model Description

Performance

Evaluation Performance

Parameters	Ascend
Model Version	NAML
Resource	Ascend 910; CPU 2.60GHz, 56cores; Memory 314G; OS Euler2.8
uploaded Date	02/23/2021 (month/day/year)
MindSpore Version	1.2.0
Dataset	MINDlarge
Training Parameters	epoch=1, steps=52869, batch_size=64, lr=0.001
Optimizer	Adam
Loss Function	Softmax Cross Entropy
outputs	probability
Speed	1pc: 62 ms/step
Total time	1pc: 54 mins

Inference Performance

Parameters	Ascend
Model Version	NAML
Resource	Ascend 910; OS Euler2.8
Uploaded Date	02/23/2021 (month/day/year)
MindSpore Version	1.2.0
Dataset	MINDlarge
batch_size	64
outputs	probability
Accuracy	AUC: 0.66

Inference on Ascend310 Performance

Parameters	Ascend
Model Version	NAML
Resource	Ascend 310
Uploaded Date	03/13/2021 (month/day/year)
MindSpore Version	1.2.0
Dataset	MINDlarge
batch_size	64
outputs	probability
Accuracy	AUC: 0.667

Description of Random Situation

In train.py, we set the seed which is used by numpy.random, mindspore.common.Initializer, mindspore.ops.composite.random_ops and mindspore.nn.probability.distribution.

ModelZoo Homepage

Please check the official homepage.