How to improve predictions?

Here is my submission for assignment 2: .
My predictions are bad, and I can’t seem to figure out what I am doing wrong. If my loss is in the range of 0-1, the prediction is still way off. Similarly, with loss of 8000 the prediction is off. What exactly is expected in terms of loss? From what I understood, the lesser the loss, the better the model.

How did you achieve loss in the range of 0-1 ? My guess is that you had included charges in the input_cols which should not be done since the model’s goal is to predict that variable.

L1 loss of 8000 is similar to what others have achieved, so you’re in a good range considering the criteria that you have to use linear regression. Min. L1 loss achieved in the community is around 4000.

Here is what you can experiment

  • Try with high LR initially where you don’t get nan, and then gradually decrease the LR .
  • children can also be considered as categorical value, can experiment with this and see if you get better results.
  • Try different loss fn
  • Can look for more data points but this is out of scope
  • Can look for better models but again out of scope w.r.t this assignment

Yes, lesser the loss better the model. But there is something you should consider that your model should not be overfitting

Hi Everyone. Hope you all doing well :slight_smile:.
After a busy schedule of my company’s work trying to complete Jovian’s fabulous Bootcamp. Now, I am sharing my Assignment-2 here -

In my case the val_loss seems to be very high, which can be improved with further training. But I will be grateful to hear few suggestions so as to improve it. :blush: :pray:
I will be starting with Assignment-3.