I have a dataset with 9912 images, what is the best way to train them based on the pixels and features of the images to be able to predict a continuous target ranging from 1 to 100? The goal is to predict the “Pawpularity” of pet photos. The Pawpularity Score is derived from each pet profile's page view statistics. In the training csv is inserted the pawpularity assigned to each image and I want to know if the image characteristics have an influence.
What is the best way to predict a continuous score based on image characteristics(pixels)?
Modified 1 year, 2 months ago
Viewed 23 times
1$\begingroup$ "Best" sounds like a matter of opinion, and is pretty vague. How will you evaluate answers? What metrics will use? Accuracy? Programmer time to write a model? Something else? $\endgroup$– D.W. ♦Mar 3, 2022 at 18:45
1$\begingroup$ Generally these ML problems are not answerable -- the only way to find out what will work best is to try things and see what works best. What approaches have you considered? What have you tried so far? $\endgroup$– D.W. ♦Mar 3, 2022 at 18:45
$\begingroup$ @D.W. I don't know, I am new at this and wanted to know how to start and what is the best way. At first I was extracting the pixels from each image and putting each image as a row and each pixel as a column to see if it had an influence on the "pawpularity. But since it gave 9912 rows and 150528 columns it wasn't loading. $\endgroup$– nunoMar 4, 2022 at 11:35
$\begingroup$ and I wanted to know the best way that when using images as input, to be able to predict a continuous target. Not to classify $\endgroup$– nunoMar 4, 2022 at 11:39
$\begingroup$ This might not be the kind of question that is a good fit for the Stack Exchange platform, then. Good luck in your work on this project! $\endgroup$– D.W. ♦Mar 4, 2022 at 19:33
| Show 1 more comment