-
Notifications
You must be signed in to change notification settings - Fork 0
Meetings
Jonathan Wallach edited this page Jun 6, 2021
·
16 revisions
Scrum master: Brendan
- finish up set up of Git Hub
- construct backlog of tasks as issues
- assign firsts tasks for sprint
- make sure everyone is clear on tasks and what needs to get done
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
- We've established that the optimal use for our prediction model will involve allowing a user to directly link to their personal Spotify
- Created a backlog of ~30 issues
- Of these issues we have placed all set-up related tasks into the first sprint
- To assign these tasks to different team members we have reviewed each members strengths/weaknesses. Jonathan and Michael are both Computer Science majors so they will be handling tasks related to research / set up of Spotify API. Brendan will be handing tasks that are related to initial data cleaning / graphing.
- Brendan's tasks are reliant on the completion of tasks by Michael and Jonathan, which is important to be aware of for the sprint.
- While discussing tasks related to storing data we've decided that we should not create an SQL database, but rather just keep our data in a CSV file
- The GitHub link is now ready to be submitted and everyone is on the same page with tasks.
Scrum master: Brendan
- check in on progress of everyone's tasks
- Identify any projected problems or changes original plan
- work on troubling tasks together
- make sure everyone is clear on what should be done by the end of the week
Jonathan Wallach
Brendan Corr
Michael Moschitto (note taker)
- Mike was able to finish up the ground work for much of the data pulling and spotipy API
- Everything looks set up on our git hub and everyone on the same page on pulling/pushing to git hub
- Figure out how to work in the genre of the song since it is not readily available on spotipy
- Jonathan took over task to figure that out while Mike and Brendan continue to work on visualizations and analysis of data
- Next steps to finish up the data exploration and make sure data is all in correct format for modeling
- If get that far, start to build model
Scrum master: Michael Moschitto
- Review current state of project
- Establish plan for moving forward
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
- The spiderweb graphs for each mood looks as expected
- Random Forest model predicted song moods with 74% accuracy
- Could be beneficial to collect feature importance values (shuffle each column, see effect)
- Going to look into training a neural network to see if we can beat the accuracy of the random forest model
- It may be too difficult to add a recommended playlist directly to a user's playlist, so we may just have to output the songs to our UI
Scrum master: Michael Moschitto
- Discuss challenges with neural network
- Discuss goal for final deliverable
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
- Original neural network is around 72% accuracy
- Struggles with indexing predictions
- Need a way for user to interact with what we create so that they can gather their song recommendations
- Still thinking a website is possible
Scrum master: Jonathan Wallach
- Update on plan for final deliverable
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
- There's a lot going on with lab/assignment deadlines and lots of remaining work to do for the write ups for the project
- As a result we may not be able to do the website
- Instead of website we can make a terminal interface
- Users can enter the mood they would like and we can output a list of songs.
- Figure out what to do with neural network results
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
- Our accuracy is not exactly as high as we'd like it to be
- Unfortunately not sure if we're realistically gonna be able to get it much higher
- This is because there is a lot of overlap in features for the different moods
- Listened to some songs to see how they were sounding
- Happy with state, think we're content with the state of predictions
Scrum master: Jonathan Wallach
- Go over current state of neural network
- Come up with a plan for handling which songs we will choose to recommend
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
- Got neural network accuracy to 77%
- Even if a song has been identified as a certain mood, there are varying levels of each mood
- To handle this we can make our model output what percentage of each mood the song fit so that we can figure out which songs are the 'most' of their mood
Scrum master: Jonathan Wallach
- Go over roles for final project
Jonathan Wallach (Note Taker)
Brendan Corr
Michael Moschitto
How are we going to format final report?
- Medium article, discussing findings and methods, but with larger emphasis on proposal for future work w/ callouts
Michael:
- Format the notebook / python files
- write about classifier / data eng
- Challenges with model accuracy, getting our own data
Jonathan:
- Future work and how we use turn this into a web/mobile application (why we didn't get to that)
- Preliminary results and discussion of what they mean within context
Brendan:
- Visualization and initial data exploration
- Explain why we believe our training set is valid (GIGO)
- Why we chose the classifiers we did <-- use stats
- DO STATS