The ML Script : Week - 5
Hello World!
Let's cut to the chase! Last week, I was stuck on the Ames Dataset (House Prediction Model).
It was really nasty, if you think like me. First, I was planning to understand all the features (input variables) of the dataset. There were approximately 80 of them. The weirdest thing was that the value indicating "None" (or the absence of a feature) was NaN. I don't understand why; the publisher could have put in a bit more effort and created a new category called "None". Maybe they just wanted to give hell to beginners! Zehahaha!!! Anyways, who cares—life is unfair.
This particular data was interesting too, despite being nasty. As you go ahead, you develop quite a liking for it. It has both nominal (no ranking among the categories) and ordinal (has some ranking or sentiment order) categorical data. We can't do the same type of encoding for both. So, you need to watch your ass.
Secondly, the real pain in the butt was the output variable (label): the sale price of the house. It has a skewed distribution (right-skewed in particular, if you really need to know). This was my first encounter with a label not being linearly distributed. So, I took some AI help to improve my own approach. It suggested taking the log of the SalePrice. This gives a normal distribution to feed into the model.
Since the model's prediction will also be in log price, this calls for some advanced stuff—i.e., a pipeline. By using a pipeline, the model automatically handles everything. Internally, it uses the log price for learning, but when asked for a prediction, it gives the actual price, not the log price.
Truly, this week was as nasty as it could get. Being stuck on a single small thing is nothing to be proud of. I was at home all week, so I was acting kinda like a Slowbro. I have advice for anyone reading this far: learning can't be done in your comfort zone.
But again, who cares. Do what you want to do! It's not my place, as I don't have any level of command over my shitty schedule to give advice! Zehahaha!
Well, see ya!