[MUSIC] We've learned now, together, we've learned
two regression models from our data, one just based on square foot of living, the other one based on these
more advanced features. So let's apply these models. Let's see what it looks like when we use
them in practice to predict house prices. So, what we're going to
do is apply this learned models to predict prices
of three houses from the data set, so three different houses. So the first house is, and
this is three that I picked that had different sizes,
different properties, they look a bit different,
in different neighborhoods. So let's just look at them. So the first house is from our data and so what we're gonna do with this house is,
so this is for not data but sales data. So out of my sales data, I'm going to select the house whose id is
equal to a particular id. And so the county record gives us string
ids for every house, it gives a number. So this one is 5309, I'm copying
from a piece of paper here, 101200. So, a particular house for
a particular id, and so I'm selecting it and if I type house1,
you'll see what it is. So it's the house that
has this particular ID. It got sold on this date for $620,000. It had four bedrooms, two and a quarter bathrooms, 2,400 square
feet of living space, and so on. Now what you can do is
in iPython notebook, you can embed not only mark ups and
texts in the Python code. But you can also embed HTML and images. So let me just do that for
you right now, just as an example. So I'm typing xm and then I am going to
add an image to my notebook and this image is an image that I downloaded from the
county records for this particular house. So, the image is on my directory,
it's called house. So, house, 53, what was the number? 09101200, so this is gonna be pretty cool. When I see a picture of the house we're
trying to make a prediction for, oops, I must have typed that wrong oh, yeah. I forgot it's .jpg. And that's pretty cool, here's
a picture of the house we're pricing. So this house was sold in 2014 for
$620,000. It has four bedrooms,
two and a quarter bathrooms. It was built, for example, in 1929. Now, let's see what our model predicts. So, We had two models. So, we had first. So the true price of the house. And just a print out to remind ourselves,
so this is house1 of price. This is 200 so if I just type print here,
it will look a little nicer. $620,000, this is the house price. Then let's see what my
first model predicts. So that was the simple square
foot model that we built. When you they it, so whenI typed .predict, the price of house1, and
it says it predicts it as $628,000, so pretty close, actually. That's really good. And let's see what adding
more features did for us. my_features_model.predict:house1, so you remember on average,
the my features model, adding more features,
gives you better performance. Now, if you look at the prediction here,
it's $720,000, so the square foot model had a better prediction
than the one with more features. So what we learned here is,
even though on average, adding more features
give a better prediction. For this one particular house, the simple model did a better job
than the more advanced model. But this is to be expected,
and not always, but sometimes gonna do better, probably works,
when average is going to do better. [MUSIC]