Children's Social Care Demand Model

This is the Python implementation of the Children's Social Care Demand Model.

It can be used as a library, as a command line tool, or as part of a web application.

It can also be used within Pyodide and has a partnering front-end application here:

https://github.com/data-to-insight/cs-demand-model-web

Principles

The model is designed to run of the SSDA903 returns, and requires several years' worth of data to be able to build a model for the system behaviour.

To create the model we take the standard SSDA903 Headers and Episodes files and merge several years of these to create a single dataset. This dataset is then analysed to give a longitudinal view of the children's experience within the system. Most notably we look at transfers between different types of placement categories (fostering, residential etc.), as well as grouping the children by age.

We then build a Stock and Flow model of the system. We use the episode start and end dates to calculate the number of children in care at any given stage (stock), and when one episode ends and another begins, we look at the type of placement the children transfer between (flow), also taking into account new children entering the system, as well as children leaving care, either returning home or ageing out of the system.

You can read more about the Data Analysis and how the data is transformed and analysed.

Technical Components

Model components:

• Configuration - How to configure the tool
• File Loader - How to load files into the tool
• Data Container - How we enrich and access the data from the model
• Predictor - takes the model and uses it to predict the number of children in care at a given point in time.

The components are designed to be re-usable and extensible.

In addition there are a couple of abstraction layers to handle loading of files in different environments.

Quickstart

Want to get started straight away? You can install from the GitHub repo and run from the command line:

pip install 'cs-demand-model[cli]'

The part after the # is optional, but will install a few extra dependencies to make the experience better when running from the command line.

You can now view the command line options by running:

demand-model

For example, you can run a quick predictive model using a sample dataset with:

demand-model predict sample://v1.zip

In this case we have used a sample dataset, but you can also use a local folder by specifying the path to the folder:

demand-model predict path/to/my/folder

The folder currently needs to have quite a specific structure with sub-folders for each year. You can make sure your folder is read correctly by running:

 demand-model list-files sample://v1.zip 

(obviously replacing the path with your own)

You can get more information about passing options to the command line tool by adding --help after the command, e.g.

demand-model analyse --help

Launching with Jupyter

You can also launch the model with Jupyter. Install the library with the jupyter extension:

pip install 'cs-demand-model[jupyter]'

Then launch Jupyter:

jupyter-lab

(or jupyter notebook if you prefer)

Launching with Jupyter-Lite (Pyodide)

Using Jupyter-Lite you can run the model in the browser without installing anything. This is a great way to get started quickly, and you can also analyse sensitive data without having to share it with 3rd parties.

We have a custom build of jupyter-lite that includes the model.

You can try it from here:

https://sfdl.org.uk/cs-demand-model-jupyter-lite/

Launching on Binder

You can also launch the model on Binder. Click to launch the sample repository on binder.