For this assignment, I decided to change my classification model in Assignment 4 to a regression model. Even though I can kind of do a regression with the confidence score from classification, I thought that maybe regression can make the prediction more precise.
First I change my data collection interface from a user input box for label entering to a slider for number input. I also changed the option from classification to regression, output from 2 to 1 because I only wanted one output value which is how much percent did I raised my hand.
I collected four groups of data. 0% for not raising my hand at all, 25% for raising a little bit, 50% for half raising, 75% for almost raising, 100% for raising my arm completely straight.
There are some issues that happened when I trained my model. For the first training, it failed. The line never went down. For the second try, it went down to halfway and stopped there. It kind of repeated this pattern back and forth between complete failure to half success.
I tried to recollect the data and increased the epochs to 100. I didn't really know what could possibly went wrong in the training process. Finally , the curve went down on my fourth or fifth try. There were still some turbulences in the curve though and it was rising at the end starting from 40 epoches. Anyway, I imported the only successful model to the third sketch.
The model kind of worked in my employ sketch. It worked well in detecting 100% and 0% but not really in the number in between.
Data Collection:editor.p5js.org/Alicelong/sketches/wSI-qXEkN