Does fastai.tabular use the dep_var to predict dep_var?

fastai
1

Does fastai.tabular use the dep_var to predict dep_var???
The predict function needs all the columns of dataframe to work (which includes the dep_var) and if you change the given dep_var to it, sometimes the returned value changes accordingly!!!
Why does it need the dep_var as part of input for the predict function?

2

When you set up your initial databunch object for training, did you include your dependent variable anywhere in your categorical or continuous variables by accident?

3

Yes, Shouldn’t I? My dep_var is a categorical variable!

4

That’s what’s happening. No, what we declare in cat_var and cont_var are the independant variables of said types so we can keep them separate.

5

My dep_var is of type integer, but it is a class label, then how should I tell fastai that its a class label

6

Hi Ramin,

You should be able to label your dep_var as “CategoryList”, fastai should then treat it as a single class label. Have a look at the relevant docs here. If your dep_var is already of type integer (not float), then I believe fastai will already default to treating it as categorical (i.e. single class). Thanks.

Yijin

7

Hi @ramin,
If your dep_var is categorical, you do not need to mention it explicitly.
But for regression problems you need to specify it when you create labels for your databunch like so :
.label_from_df(cols=dep_var, label_cls=FloatList)

I think lesson 6 covers the topic in details.
Hope this helped