AnnotateChange

Welcome to the repository of the "AnnotateChange" application. This application was created to collect annotations of time series data in order to construct the Turing Change Point Dataset (TCPD). The TCPD is a dataset of real-world time series used to evaluate change point detection algorithms. For the change point detection benchmark that was created using this dataset, see the Turing Change Point Detection Benchmark repository.

Any work that uses this repository should cite our paper: Van den Burg & Williams - An Evaluation of Change Point Detection Algorithms (2020). You can use the following BibTeX entry:

@article{vandenburg2020evaluation,
        title={An Evaluation of Change Point Detection Algorithms},
        author={{Van den Burg}, G. J. J. and Williams, C. K. I.},
        journal={arXiv preprint arXiv:2003.06222},
        year={2020}
}

Here's a screenshot of what the application looks like during the annotation process:

Some of the features of AnnotateChange include:

Admin panel to add/remove datasets, add/remove annotation tasks, add/remove users, and inspect incoming annotations.
Basic user management: authentication, email confirmation, forgotten password, automatic log out after inactivity, etc. Users are only allowed to register using an email address from an approved domain.
Task assignment of time series to user is done on the fly, ensuring no user ever annotates the same dataset twice, and prioritising datasets that are close to a desired number of annotations.
Interactive graph of a time series that supports pan and zoom, support for multidimensional time series.
Mandatory "demo" to onboard the user to change point annotation.
Backup of annotations to the admin via email.
Time series datasets are verified upon upload acccording to a strict schema.

Getting Started

Below are instructions for setting up the application for local development and for running the application with Docker.

Basic

AnnotateChange can be launched quickly for local development as follows:

Clone the repo

$ git clone https://github.com/alan-turing-institute/AnnotateChange
$ cd AnnotateChange

Set up a virtual environment and install dependencies (requires Python 3.7+)

$ sudo apt-get install -y python3-venv # assuming Ubuntu
$ pip install wheel
$ python3 -m venv ./venv
$ source ./venv/bin/activate
$ pip install -r requirements.txt

Create local development environment file
```
$ cp .env.example .env.development
$ sed -i 's/DB_TYPE=mysql/DB_TYPE=sqlite3/g' .env.development
```
With DB_TYPE=sqlite3, we don't have to deal with MySQL locally.
Initialize the database (this will be a local app.db file).
```
$ ./flask.sh db upgrade
```
Create the admin user account
```
$ ./flask.sh admin add --auto-confirm-email
```
The --auto-confirm-email flag automatically marks the email address of the admin user as confirmed. This is mostly useful in development environments when you don't have a mail address set up yet.
Run the application
```
$ ./flask.sh run
```
This should tell you where its running, probably localhost:5000. You should be able to log in with the admin account you've just created.
As admin, upload ALL demo datasets (included in demo_data) through: Admin Panel -> Add dataset. You should then be able to follow the introduction to the app (available from the landing page).
After completing the instruction, you then will be able to access the user interface ("Home") to annotate your own time series.

Docker

To use AnnotateChange locally using Docker, follow the steps below. For a full-fledged installation on a server, see the deployment instructions.

Install docker and docker-compose.

Clone this repository and switch to it:

$ git clone https://github.com/alan-turing-institute/AnnotateChange
$ cd AnnotateChange

Build the docker image:

$ docker build -t gjjvdburg/annotatechange .

Create the directory for persistent MySQL database storage:

$ mkdir -p persist/{instance,mysql}
$ sudo chown :1024 persist/instance
$ chmod 775 persist/instance
$ chmod g+s persist/instance

Copy the environment variables file:
```
$ cp .env.example .env
```
Some environment variables can be adjusted if needed. For example, when moving to production, you'll need to change the FLASK_ENV variable accordingly. Please also make sure to set a proper SECRET_KEY and AC_MYSQL_PASSWORD (= MYSQL_PASSWORD). You'll also need to configure a mail account so the application can send out emails for registration etc. This is what the variables prefixed with MAIL_ are for. The ADMIN_EMAIL is likely your own email, it is used when the app encounters an error and to send backups of the annotation records. You can limit the email domains users can use with the USER_EMAIL_DOMAINS variable. See the config.py file for more info on the configuration options.
Create a local docker network for communiation between the AnnotateChange app and the MySQL server:
```
$ docker network create web
```
Launch the services with docker-compose
```
$ docker-compose up
```
You may need to wait 2 minutes here before the database is initialized. If all goes well, you should be able to point your browser to localhost:7831 and see the landing page of the application. Stop the service before continuing to the next step (by pressing Ctrl+C).
Once you have the app running, you'll want to create an admin account so you can upload datasets, manage tasks and users, and download annotation results. This can be done using the following command:
```
$ docker-compose run --entrypoint 'flask admin add --auto-confirm-email' annotatechange
```
As admin, upload ALL demo datasets (included in demo_data) through: Admin Panel -> Add dataset. You should then be able to follow the introduction to the app (available from the landing page).
After completing the instruction, you then will be able to access the user interface ("Home") to annotate your own time series.

Notes

This codebase is provided "as is". If you find any problems, please raise an issue on GitHub.

The code is licensed under the MIT License.

This code was written by Gertjan van den Burg with helpful comments provided by Chris Williams.

Some implementation details

Below are some thoughts that may help make sense of the codebase.