PyTorch Linear regression to predict fish weight

Hello, I have tried to predict the fish weight with Linear regression
but the results is not very good and I have some question:
1- the loss some times starts very high numbers (is that because the model initialize the training with random W and B?
2- does linear regression works for this problem ?
3- any other data that I can use it for linear regression?

the notebook is in the link

1-Yes, the model starts with random weights and bias, so the first loss can be very large.
2-Yes, the real question is: “Linear regression is the best way to solve this problem?”
3-In this link you’ll get a lot of good datasets, https://www.kaggle.com/rtatman/datasets-for-regression-analysis

1 Like

Use l1 loss or train for more epochs.
You have to use any weights and biases so why not normally distributed weights and biases.

1 Like