The Safer Streets Priority Finder (SSPF) enables you to analyze the risk to bicyclists and pedestrians on your community’s roads. You can use your local road, crash, and study area data or select from nationally available datasets to:
- Explore descriptive statistics related to your crash data
- Develop a Sliding Windows Analysis using historical crash data to inform a High Injury Network
- Develop a Safer Streets Model to estimate risk along your road network, even in areas that haven't had any reported crashes recently
Check out the Safer Streets Priority Finder Tool here.
You don't need to follow these instructions to use the tool, which is available for public use at the link included above. You only need to follow these instructions if you'd like to build your own version of the SSPF software.
-
Linux - You'll need sudo access on a Linux command line.
-
A PostgreSQL Relational Database - Access to a PostgreSQL database that follows the schema and table structure provided here.
-
Static tables - Upload the following table into the 'static' schema on the PostgreSQL datatable.
-
Fatality Analysis Reporting System (FARS) 2015 - 2019 - as, static.fars_processed
-
USDOT-OST / Pedestrian-Fatality-Risk-Project - as, static.national_tracts
-
US Census Counties - as, static.us_county_2018
-
National Open Street Map Roads Dataset, Available from GeoFabrik - as, static.osm_centerlines
-
Functional Classification Priors - as, static.national_fclass_priors
Datasets 1-4 need a state and county Federal Information Processing Standards (FIPS) code, where the state FIPS code is always two digits in length, and county codes are three digits, including leading zeros as needed. FIPS codes on each dataset should be stored in TEXT or VARCHAR format.
See this file for more information on the data structure for each table listed above.
-
-
Complete system variables - Fill out your system variables, so the Docker container can link the user information with the data. You'll need to do this here and here. Do not relocate this file.
-
Docker - If you don't already have Docker installed on your machine, you can get started here.
The easiest way to launch the application is setup a PostgreSQL database following the structure provided above. Once that's ready, launch a Docker container with the model processor, and a seperate container with the SSPF Tool.
sudo up
git clone https://github.com/tooledesign/Safer-Streets-Priority-Finder.git
Navigate to the first file where system variables are required:
cd Safer-Streets-Priority-Finder/safer_streets_priority_finder/
You can update the system variables on the CL with nano text editor.
nano env_variables.R
Update all variables in that file. Here's a breakdown of what each variable refers to:
- SSPF_EMAIL_PASSWORD - password to an email account that sends a notification to users that a model has started or finished
- SSPF_EMAIL_ADDRESS - email address to send notification
- SSPF_AMAZON_HOST_ADDRESS* - host address to PostgreSQL Relational Database
- SSPF_AMAZON_PASSWORD - password to PostgreSQL Relational Database
- SSPF_AMAZON_USERNAME - username for PostgreSQL Relational Database
- SSPF_AMAZON_DATABASE - the name of the database with schema structure for the tool
- AWS_ACCESS_KEY_ID - AWS S3 access key
- AWS_SECRET_ACCESS_KEY - AWS S3 secret access key
- AWS_DEFAULT_REGION - AWS S3 location
- S3_BUCKET - Name of the S3 bucket
- PEPPER - random string of character, used in encyption
- USER_DATA_KEY - key to decipher encrypted messages
Now do the same for the report processor.
cd ..
cd Safer-Streets-Priority-Finder/report_processor/
nano env_variables.R
And the same for the model processor. Note, the model processor does not need variables to access the AWS S3 bucket.
cd ..
cd Safer-Streets-Priority-Finder/vulusr_model_processor/
nano env_variables.R
Note: by default, Docker runs on root. Either make sure you are logged in as root by using sudo su, or prepend all of the following Docker commands with sudo.
First, make sure you're logged into your Docker account.
docker login
If you haven't already done so, you'll need a network to link systems together.
docker network create sp-example-net
Ensure you the variable configuration has been updated for this process.
Navigate to the Safer-Streets-Priority-Finder/vulusr_model_processor/ directory. Then build the model processor image with the following:
docker build -t sspf_model_processor .
This will build an image called sspf_model_processor.
If you plan to store this information in a Docker Hub, you'll want to namespace the image with,
tag sspf_report_processor [yourDockerAccountNameHere]/sspf_report_processor:[version_number]
After that's complete, you launch a container with,
docker run -d -v /var/run/docker.sock:/var/run/docker.sock --net sp-example-net --name=sspf_model_processor --restart=always -p 9001:3839 sspf_model_processor
Verify the model processor is now running:
docker ps
You should not see any indication the model processor is restarting. To check for any errors, use the following:
docker logs sspf_model_processor
At this time, the model processor is running in a detached Docker container listening for a user to request a model. Results from the model processor are stored in the PostgreSQL database identified in the env_variables.R file. Once a container has finished building a model, the user will be notified by email. At this time, that container will be marked stale using the gen_management.docker_status table. The developers found that a single containerized model processor can't continuously build Bayesian models with RStan. We recommend maintaining this structure.
See the "Maintaining the model processor manually" section below to maintain the model processor manually.
OR
See the "Maintaining the model processor automatically" section below to maintain the model processor automatically.
Ensure the variable configuration has been updated for this process.
Navigate to the Safer-Streets-Priority-Finder/vulusr_model_processor/ directory, then build the model processor image with the following:
docker build -t sspf_report_processor .
This will build an image called, sspf_report_processor on your local machine.
If you plan to store this information in a Docker Hub, you'll want to namespace the image with,
tag sspf_report_processor [yourDockerAccountNameHere]/sspf_report_processor:[version_number]
To launch a container, use the following command:
docker run -d -v /var/run/docker.sock:/var/run/docker.sock --net sp-example-net --name=sspf_report_processor --restart=always -p 25250:25250 sspf_report_processor
Verify the report processor is now running:
docker ps
You should not see any indication the report processor is restarting. To check for any errors, use the following:
docker logs sspf_report_processor
At this time, the report processor is listening for a user to request a report. Reports are processed one at a time. If multiple reports are requested at the same time, the processor will start with the study with the earliest start timestamp. A status message will appear in the SSPF UI to notify the user of the processor's step. Reports are stored in an AWS S3 bucket. The user will be notified by email when a report is ready.
Navigate to the safer-streets-priority-finder directory. This directory should container a Dockerfile.
cd Safer-Streets-Priority-Finder/safer_streets_priority_finder/
Then build the Safer Streets Priority Finder with the following:
docker build -t sspf .
If you want to save your image on Docker Hub,
tag sspf_report_processor [yourDockerAccountNameHere]/sspf_report_processor:[version_number]
docker run -d -v /var/run/docker.sock:/var/run/docker.sock --net sp-example-net --name=sspf --restart=always -p 9000:3838 sspf
Verify the Safer Streets Priority Finder is now running:
docker ps
You should not see any indication the Safer Streets Priority Finder is restarting. To check for any errors, use the following:
docker logs sspf
Now navigate to http://localhost:9000/, and you should see the Safer Streets Priority Finder.
The model processor is designed to run one time only before needing to have the Docker container refreshed. This process can be done manually via the following steps, or can be automated (details below following the manual process).
Run the following command to stop the Docker container that is running the model processor.
docker stop sspf_model_processor
Run the following commmand to remove the old Docker container.
docker container prune -f
Run the following SQL query from a SQL editor (psql, DBeaver, PGAdmin, etc.) to update the Docker container status flag.
UPDATE gen_management.docker_status SET is_fresh = TRUE;
Run the following command to spawn a new Docker container running the SSPF model processor.
docker run -d -v /var/run/docker.sock:/var/run/docker.sock --net sp-example-net --name=sspf_model_processor --restart=always -p 9001:3839 sspf_model_processor
If you want to automate the process outlined in the above section, to avoid having to manually reset the Docker container each time you run a model, use the dockerManager.py script.
dockerManager.py is a python tool that automatically refreshes the model processor Docker container after successfully running a model, and is run by a linux cron job. The following instructions detail how to set up and use this tool.
psycopg2 is a python module required for communicating with the PostgreSQL database. If you have already installed PostgreSQL locally on the same machine on which you are running the model processor, then this requirement is already satisfied. If you are using a remote PostgreSQL server, separate from the model processor machine, run the following commands to install psycopg2 without having to fully install PostgreSQL.
sudo apt-get install libpq-dev
sudo su
pip3 install psycopg2
exit
dockerManager.py requires connecting to the PostgreSQL database, and so needs the connection credentials. Store these credentials in a configparser file located at ./vulusr_model_processor/python/.sspf_config. Do the following to set up the config file (a template file should already exist there):
nano ./vulusr_model_processor/python/.sspf_config
Enter the following text:
[postgres]
host=%hostname%
dbname=%databasename%
user=%username%
password=%password%
Substitute in the connection values you used when you set up the PostgreSQL database earlier. When done, exit nano and save the file.
dockerManager.py relies on a linux cron job to run the python script every X minutes, typically every 1 minute. Do the following to set up the cron job information.
Edit the root-level cron jobs:
sudo crontab -e
Add a line for controlling the dockerManager.py tool.
* * * * * python3 path/to/vulusr_model_processor/directory/python/dockerManager.py
Save and exit.
By default, dockerManager.py should be stored at ./vulusr_model_processor/python/dockerManager.py. If you store it elsewhere, be sure to update the filepath accordingly in the previous step when editing the crontab file. The full file path must be entered in the crontab file.
Similar to the above automated process, this repository includes a tool (userDataCleanup.py) that automatically removes old user data from the PostgreSQL server from previous model runs, if the user has been inactive for at least 6 months. If you wish to implement this functionality, follow a similar process to set up a cronjob to regularly run the tool.
Follow steps 1 and 2 from the above process for dockerManager.py if you have not already.
userDataCleanup.py can run less frequently. As with step 3 above, add a line to the root-level crontab file. e.g. to run userDataCleanup.py once weekly on Saturdays at 2am, add the following line to the root-level crontab file:
0 2 * * 6 python3 path/to/vulusr_model_processor/directory/python/userDataCleanup.py
If you want to run the UI from your local RStudio environment, here's a list of libraries that you'll need. Open this project within RStudio. The Docker container uses this very specific work directory, that you'll need to comment out. After that, run this file and that should launch the UI on your local computer.
If you want customized support implementing this tool in the AWS cloud, or need support with Docker containerization, please get in touch with us. You can send questions or support inquiries to the developers by emailing, [email protected].