I feel like I’m missing something with ML.NET.

As a POC before investing serious computer time. I tried creating a CSV file, 2 columns with 1000s of rows:

Greeting,Reply
My name is Jimmy,Hi Jimmy! How are you today?
My name is Sally,Hi Sally! How are you today?
My name is Mike,Hi Mike! How are you today?

The Model Builder has classification models (I don’t believe this is a classification model because I want the thing to be able to predict something new so when it received “My name is Poutine” it will say “Hi Poutine! How are you today?”

I tried anyway and kept getting 0% success on the model.

I came up with another idea, to convert the words to numeric values and then return an array of numeric values.

Doesn’t seem like ML.NET can return arrays very easily. I kept getting errors about invalid schema expecting single Key returns rather than arrays.

When I followed up with Bing AI on this it suggested:

Unfortunately, ML.NET does not currently have built-in support for text generation or sequence-to-sequence models. You may want to consider using a different machine learning framework that supports these capabilities, such as TensorFlow or PyTorch.

Am I getting accurate info from the AI? Should I be looking outside of the MicrosoftSphere for this stuff?

Anything you’d recommend?

  • nibblebit@programming.devM
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    Hey, I’ve worked with ML.NET before and it’s not the best framework for C#, but it is capable. I’m having trouble understanding what the goal of your model would be. Is it just text prediction or classification? ML.NET and pretty much any ML framework do need some experience with machine learning methods and models to achieve good results. Is this your first time doing something with ML?

    I would look into the SciSharp stack a great collection of AI/ML framework bindings for common frameworks I would recommend looking into Torch.Net and Keras in particular.

    • TurianHammer@lemm.eeOP
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      No I’ve done stuff with pure classification before where the answer is 1 of X possibilities.

      I’ve also done some stuff with numeric value prediction but I’m hoping to try some things with text.