This repo includes the web scraping, data preparation, and model building code that I used to create a rainbow predictor for the Seattle area. Please see my repo at https://github.com/marybarnes37/rainbow_flask/ for information on the final implementation.
Who doesn’t take joy at the sight of a rainbow, or in pointing one out to others in their vicinity?
This project builds out two of the components that would be required to create a ‘Rainbow Alert’ phone app. The full version of the app has three components. First, it predicts the probability of witnessing a rainbow in the user’s geographic area on that day (a morning prediction and an afternoon prediction). Second, it offers a real-time probability of spotting a rainbow which is updated at intervals throughout the day (as long as the sun is below 42 degrees). Third, if a user spots a rainbow, takes a picture of it and uploads it to the app, the app will verify that the photo contains a rainbow and then send a “rainbow alert’ notification to other users in the geographic area. The purpose of validating the photos with an image classifier is to ensure that a user cannot set off an erroneous rainbow alert in their area.
The code in this repo tackles the first data science elements of the full Rainbow Alert app: it returns the‘probability of rainbow given weather conditions and time of day’ in Seattle.
Should I decide not to eventually build the fully-functioning app, a predictive model for rainbows could still be a welcome addition to a standard weather forecast phone app. Cloudy with a chance of rainbows!
The instagram_sraper folder includes code for gathering location and timestamp information from Instagram posts with a given hashtag. I did not end up using this data because I was not confident in making sense of Instagrams timestamp format.
The flickrsraper folder includes code for gathering location and timestamp information from Flickr posts with a given tag from the Flickr API. I used weather data from these timestamps to train my model.
The metar folder contains data files that list METAR weather stations. This data could be used to extend the rainbow predictor to areas outside of Seattle. It was not necessary for the Seattle-only implementation.
I was given access to a non-public Weather API that allowed me to pull in weather data from specified times.
The flickrsraper folder contains all of the data preparation files for the Seattle implementation.
In a Jupyter Notebook, I fit a logistic regression, a gradient boosted decision tree, and a random forest and use cross validation to choose the best.
The pickles folder contains pickled dictionaries and trained models that are backups. They were originally created in a Jupyter notebook.
The Seattle alert system is live at infinityrainbows.com.
In the future, I would like to set up Twilio alerts so that users can receive SMS notifications.
If I am also able to build the rainbow image classifier, I would like to come up with an app that notifies users when there is a confirmed rainbow sighting in their area.