I'm taking a first swing at training a Core ML model using Create ML. I'm feeling a little out of my depth on a couple of first principles, so I'm hoping someone has some good advice.
Here's my task — I have user input for a drawing that is basically an Array<CGPoint> representing the user's touch locations in series. I want to be able to classify the drawing from a limited number of shapes (let's say "ellipse", "triangle", "rectangle").
In the Create ML app, I'm having difficulty figuring out how to provide my training data in a format it likes or will understand. As it stands right now, I have a CSV file representing an individual drawing, that is just the exported CGPoints, with X and Y as columns and each row representing a point. I have a file structure with folders labeled "ellipse", "triangle", "rectangle", each containing a number of CSV files representing individual drawings.
It seems like Create ML, to work with tabular data, wants me to provide a target (like "triangle") for each row. But that doesn't make sense to me because these values only represent "triangle" when taken in aggregate.
So I started looking at providing tabular data where the number of dimensions is greater than 2, and kind of came up short. I see that the CoreML library has some representation of multi-dimensional arrays, but I'm feeling like I must be missing some really obvious thing that would allow me to train a model using Create ML.
Does anybody have any suggestions on how to shape my data so that Create ML can ingest and train on it? I can move the data into any format/shape required. How can I take a series of x and y points and tell Create ML "this one's a triangle, this one's an ellipse, etc..."
Thanks!
Here's my task — I have user input for a drawing that is basically an Array<CGPoint> representing the user's touch locations in series. I want to be able to classify the drawing from a limited number of shapes (let's say "ellipse", "triangle", "rectangle").
In the Create ML app, I'm having difficulty figuring out how to provide my training data in a format it likes or will understand. As it stands right now, I have a CSV file representing an individual drawing, that is just the exported CGPoints, with X and Y as columns and each row representing a point. I have a file structure with folders labeled "ellipse", "triangle", "rectangle", each containing a number of CSV files representing individual drawings.
It seems like Create ML, to work with tabular data, wants me to provide a target (like "triangle") for each row. But that doesn't make sense to me because these values only represent "triangle" when taken in aggregate.
So I started looking at providing tabular data where the number of dimensions is greater than 2, and kind of came up short. I see that the CoreML library has some representation of multi-dimensional arrays, but I'm feeling like I must be missing some really obvious thing that would allow me to train a model using Create ML.
Does anybody have any suggestions on how to shape my data so that Create ML can ingest and train on it? I can move the data into any format/shape required. How can I take a series of x and y points and tell Create ML "this one's a triangle, this one's an ellipse, etc..."
Thanks!