Bunching Delays Prediction module.

To run without the docker container:

1. Download the file datapoints_all.pkl (available in Dittlab, distributed as necessary), and put it into the sampledata folder.

2. Make sure MySQL server is running on your host. If necessary, change the appropriate lines in the configuration in the putSampleDataIntoMySQL.py file:

db = mysql.connect(
    host="localhost",
    user="...",
    passwd="...",
    database="mytrac"
)

3. Run putSampleDataIntoMySQL.py. The sample data is now in your MySQL db in the necessary format.

4. Run startServer.py.

5. Your delay prediction server is now running. You can verify it by navigating your browser to

http://localhost:5000/

You should see the following message:

Delay Prediction Server is running!

Usage documentation

When the server starts, it automatically checks the database and finds all unique values of linedir - an identification of a particular PT route (line number and direction combined). Then it reads the datapoints for each linedir, creates, trains and saves to disk the trained ML model.

If, for some reason, you want to rerun the training for a particular linedir, you can use the following command:

http://localhost:5000/training?parameter1=value1&parameter2=value2&...

The following parameters (all of them optional) can be submitted to the training request:

linedir : Unique linedir for which to perform the re-training, e.g. "13-1"

from_date : Starting date from which to use data for training, in the format "yyyymmdd" e.g. "00000000" (for all days), "20150301"

to_date : End date before which to use data for training, in the format "yyyymmdd" e.g. "30001231" (for all days), "20150324"

The main usage of the module is for prediction requests. Each such request can provide you with a prediction of the duration and the type (cluster) of the existing bunching swings formation.

The request can be sent in the following form:

http://localhost:5000/prediction?parameter1=value1&parameter2=value2&...

The following parameters are accepted:

linedir : Unique linedir for which to perform the re-training, e.g. "13-1"

Date and time of the moment at which to perform the prediction. If absent, the current date and time will be used (normal mode of operations), but earlier dates or times can be used for testing or other purposes.

date in the format "yyyymmdd" e.g. "20150327"

time in the format of the number of seconds since the beginning of the day, e.g. 46971 for 13h 2m 51s

The prediction's response will be in the following format, where cluster represents the type of cluster (as described in the Deliverable 3.2), and the duration represents the predicted duration in seconds:

{"cluster": "3", "duration": "3818"}

If no bunching is observed, both values will be set to -1.

Examples of prediction requests:

http://localhost:5000/prediction?linedir=1-0&date=20150303&time=36800


http://localhost:5000/prediction?linedir=1-0&date=20150303&time=46800


http://localhost:5000/prediction?linedir=1-0&date=20150301&time=46800