Daply

[rodrygoes @rodrygoesdotfun]

Day 19 of becoming cracked in ML

Been away for a few days now, ig I'm back to learning.

I revised the models I've learnt so far, then went back into my California Housing Linear Regression project(i felt like i was seeing everything for the first time again, icel).

I went through the data loading, Exploratory Data Analysis, preparation and train/test split, then got into feature scaling.

The idea is simple: if one feature ranges around "5" while another ranges around "3,000", Gradient Descent can have a much harder time navigating the loss surface efficiently.

So I learned how standardization works:

"x_scaled = (x - μ) / σ"

I also learnt a criticaltip: Don't scale your entire dataset before the train/test split, otherwise, information from the test set can leak into the training process which can mislead the evaluation.

I then trained Linear Regression from scratch using Gradient Descent and got introduced to Batch, Stochastic and Mini-Batch Gradient Descent.

Stochastic and Mini-Batch are still a little blurry but still have similar syntax to perceptrons, so I'm going to focus on Batch Gradient Descent for now and come back to them when I have a better foundation.

Finally, I started evaluating the model using MSE, RMSE and MAE, with model interpretation and a comparison against Scikit-Learn still left to do.

I'm roughly 90–95% done with the project, its not been easy and it's also not been hard, buh we moveee!!!

I plan on moving to the next model tomorrow and just try to complete this project alongside.

[Daply @_Daply]

@mark_k Is elon a time traveller?

[Daply @_Daply]

good morning world!