https://www.kaggle.com/willkoehrsen/introduction-to-manual-feature-engineering
I thought I should make a note of what I was doing. Qiita says that memos are OK, so let's make a note here.
It's complicated to think that this notebook combines various tables, but
Just group by with client_id and count, and add that line to application_train to increase the feature amount, It's easy to think of it.
However, I feel that this KDE plot is useful in many places. Of course I've put out this kind of graph, but it's nice to have a name. For a common understanding.
Flattening the multi-level index to one level seems to be useful somewhere other than machine learning.
The continuation Function for Numeric Aggregations https://www.kaggle.com/willkoehrsen/introduction-to-manual-feature-engineering#Function-for-Numeric-Aggregations From.
If you pass the list, it will sort.
sorted([1,2,4,4,1,1,2,5,555,230])
[1, 1, 1, 2, 2, 4, 4, 5, 230, 555]
Functions can also be used for sorting, so new_corrs = sorted(new_corrs, key = lambda x: abs(x[1]), reverse = True) Sort by absolute value of x 1 (in reverse order).
--institution institution --handle with A A ―― by pure chance Just by chance, just by chance
I just remembered it or forgot it.
Recommended Posts