Time-Series-Prediction

This repository contains my implementations for time series prediction task.

Requirements

Python 3.6
PyTorch 1.6.0
Tensorboard 1.15.0
XGBoost 1.2.0
Sklearn 0.19.1
Numpy 1.16.2
Tqdm 4.49.0

Introduction

Project

The project structure is below:

.
├── PyTorch
│   ├── LSTNet
│   │   ├── test.py
│   │   ├── layers.py
│   │   └── train.py
│   └── utils
│       ├── param_parser.py
│       └── data_helpers.py
├── data
│   ├── Train / Validation /Test_sample.json
│   ├── Train / Validation / Test_BOW_sample.json
│   └── Train / Validation / Test_pairwise_sample.json
├── LICENSE
├── README.md
└── requirements.txt

Data

See data format in /data folder which including the data sample files. For example, train_sample.json is like:

{"id": "6", "content": ["year", "ruined", "summer", "vacation-a", "two-week", "vacation", "wife", "family", "cabin", "lake", "northern", "ontario", "located", "boundary", "canada-by", "bringing", "modern", "convenience", "wa", "convenient", "good", "ipad", "admiring", "beauty", "nature", "checked", "e-mail", "paddling", "canoe", "twitter", "feed", "devouring", "great", "amusing", "stuck", "workday", "diet", "newspaper", "morning", "wa", "problem", "wa", "behaving", "office", "sticking", "unending", "news", "cycle", "body", "wa", "vacation", "head", "wasnt", "year", "made", "mind", "social", "medium", "experiment", "reverse", "withdrawal", "internet", "manage", "unplug", "knew", "wouldnt", "easy", "im", "good", "self-denial", "wa", "determined", "started", "physical", "restraint", "handing", "ipad", "wife", "helpfully", "announced", "wa", "read", "book", "club", "inclined", "relinquish", "tablet", "moment", "stroke", "luck", "cell", "phone", "signal", "canadian", "cabin", "wa", "spottier", "past", "making", "attempt", "cheating", "experience", "frustration", "wa", "trapped", "forced", "comply", "good", "intention", "largely", "cut", "e-mail", "twitter", "favorite", "newspaper", "website", "connect", "world", "radio-and", "radio", "listen", "choice", "planned", "read", "book", "experienced", "criminal", "plot", "street", "los", "angeles", "cutthroat", "battle", "cancer", "lab", "psyche", "london", "social", "butterfly", "magazine", "read", "im", "claiming", "cut", "internet", "completely", "day", "biked", "nearest", "town", "reward", "sat", "park", "bench", "front", "public", "library", "wi-fi", "back", "cabin", "suffered", "slow", "dial-up", "connection", "day", "check", "e-mail", "tale", "self-denial", "ha", "happy", "ending-for", "determination", "deep", "breathing", "strong", "support", "wife", "succeeded", "vacation", "struggle", "internet", "realizing", "finally", "wa", "ipad", "wa", "problem", "knew", "passed", "starbucks", "wife", "asked", "wanted", "stop", "wi-fi", "dont", "sound", "pleased", "return", "post-vacation", "situation", "test", "begin", "stay", "wagon", "im", "back", "work", "time", "compulsion", "whats", "overwhelming", "crucial", "livelihood", "intention", "giving", "membership", "cult", "immediacy", "hope", "resist", "temptation", "reflexively", "check", "e-mail", "minute", "lead", "long", "im", "checking", "twitter", "feed", "website", "vacation", "supposed", "reset", "brain", "productive", "hoping", "worked"], "question": ["doe", "underlined", "word", "restraint"], "pos_text": ["calm", "controlled", "behavior"], "neg_text": ["relaxing", "move", "strong", "determination", "unshakable", "faith"], "diff": 0.550373134328}

"id": just the id.
"content": the word segment of the content.
"question": The word segment of the question.
"pos_text": The word segment of the correct option.
"neg_text": The word segment of the wrong options.
"diff": The difficulty of the question.

Data Format

This repository can be used in other similiar datasets in two ways:

Modify your datasets into the same format of the sample.
Modify the data preprocessing code in data_helpers.py.

Anyway, it should depend on what your data and task are.

Usage

See Usage

Network Structure

The framework of LSTNet:

Reference

If you want to follow the paper or utilize the code, please note the following info in your work:

Model LSTNet

@inproceedings{lai2018modeling,
  title={Modeling long-and short-term temporal patterns with deep neural networks},
  author={Lai, Guokun and Chang, Wei-Cheng and Yang, Yiming and Liu, Hanxiao},
  booktitle={The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval},
  pages={95--104},
  year={2018}
}

About Me

黄威，Randolph

SCU SE Bachelor; USTC CS Ph.D.

Email: chinawolfman@hotmail.com

My Blog: randolph.pro

LinkedIn: randolph's linkedin

RandolphVI / Time-Series-Prediction