User Stories

[ang-zeyu]

App

• I should be able to interface with BLE functionalities on CC2650, so that I can gather data for training and prediction
• I should be able to navigate to 2 separate screens for training and prediction (autoplay), so that the separation of functionalities is clear.
• I should be able to initiate and stop motion (running, walking, ...) training at any time with a button
  • I should be able to sample and collect motion data into a temporary store
  • The activity should be selectable as a dropdown / secondary popup / etc. @Florence-100
  • I should be able to stop training with the same button, and call the motion training API (API not done) @ang-zeyu
• I should be able to source training and autoplay files from a folder on my phone
• I should be able to populate song moods of training music files with a hardcoded .json in the app
• I should be able to populate song moods of (prediction) music files with an API call
• I should be able to see song moods of loaded songs in the UI of
  • training songs
  • prediction songs
• I should be able to have a UI to call an API for POST-ing song title -> mood mapping
• I should be able to initiate song training (play randomised songs from folder) at any time with a button
  • I should be able to have the sensor collect data for a while at the beginning of each song playing (3-5s) (done)
  • When I skip a song before it has been completed, I will send a training API call with an "inverted" flag that inverts the moods
  • When I let a song play to completion, I will send a normal training API call
• I should be able to initiate song autoplay mode using sensor IoT data on the prediction screen with a button
  • I should be able to have the sensor collect data for a while at the beginning of each song playing (3-5s) (half done)
  • So that I can call the prediction API to get a song mood recommendation
• irrelevant - no longer labels I should be able to populate song labels of music files with an API call

Backend

• I should be able to administrate EC2 instances, for serving the backend APIs and training the ML models
• I should support a motion training MQTT API that takes sensor inputs over some period + the activity
• I should support a song training MQTT API that takes sensor inputs over a short period + the song mood
• I should support a song prediction API that takes sensor inputs over a short period, and recommend a song mood
• I should be able to store song moods of music files with an API call
• I should be able to retrieve song moods of music files with an API call
• !! I should be able to save training data to some local file / database, so that I can tweak and adjust the model anytime during development
  • Correspondingly, I should be able to retrieving said training data and train ML models at any time
• irrelevant - no longer labels I should support a song labels API that takes a list of song titles and artists as inputs (ideally) and returns their corresponding song labels
• irrelevant - no longer labels I should be able to interface with the millionsongdataset.com, and possibly others, so that I can retrieve song labels

CC2650

• We should calibrate our sensors if needed, so that aggregating data is consistent (not done)