3 ms·

So, I am a total ML noob. The thing I haven't found a straight answer to is, what is a model. I mean when it is in production? Is it just some random blob that

• Mentlo 5y ago
  Depends on how it's put into production, but you can deploy a model as a RESTful API that has a defined interface and a defined output. What it does underneath is less important to you I guess. So for all intents and purposes, yes, a model in production is something you feed a predefined set of data points and it gives you a predefined format of output.
• jazzyjackson 5y ago
  So, you know how a straight line is defined as mx + b, where you just have two parameters: slope and intercept ? Your input value is X, you multiply it by your slope and add your intercept to get the output (the Y value on the line). The 'training' of an ML algo is really just finding the line-of-best-fit so that you can make predictions. So your line-of-best-fit is encoded in these two parameters, allowing you to make predictions about what the output would be for arbitrary input. The problems people are throwing at ML have many more parameters and dimensions, but the training is a matter of finding those parameters that come closest to predicting the outcome. The 'model' is this set of parameters that allows the function to make predictions. (disclaimer: also an ML noob, correct me if I'm wrong)
  • rdevsrex 5y ago
    thanks!