Let's decide the lecture of PyCon JP 2016 by combinatorial optimization

Let's decide the lecture of PyCon JP 2016

(This is just a joke)

This year's theme for PyCon is "Everyone is different, everyone is good".

So, let's choose a lecture that incorporates all the elements. Use the set cover problem in Combinatorial Optimization (http://qiita.com/Tsutomu-KKE@github/items/bfbf4c185ed7004b5721).

Create lecture candidates at random

Randomly create the nationality of the speaker, the level of the lecture, the field of the lecture, etc. The score is, for example, the number of likes.

python


import numpy as np, pandas as pd
np.random.seed(1)
n = 20
a = pd.DataFrame({
        'Country': np.random.choice('アメリカ イギリス インド フランス ロシア 中Country'.split(), n),
        'level': np.random.choice('Beginner Intermediate Advanced'.split(), n),
        'Field': np.random.choice('Data Analysis Optimization Machine Learning Document Web'.split(), n),
        'score': np.random.randint(1, 10, n),
    })
print(a)
level Field Country score
0 Beginner Web China
1 Advanced optimisation France
2 Beginner Data analysis Russia
... ... ... ...
19 Beginner Data analysis France

Solve the set cover problem

Use ortoolpy.set_covering. Since it is minimized, the reciprocal of the score is weighted.

python


from ortoolpy import set_covering
n = sum(b.nunique() for b in [a.level, a.Field, a.Country]) #Total unique number
res = set_covering(n, [(1/r.score, r[:3].tolist()) for _, r in a.iterrows()])
print(a.ix[res])
level Field Country score
5 Advanced Machine learning France
8 Intermediate document America
10 Beginner optimisation Russia
13 Intermediate Data analysis England
14 Intermediate optimisation India
16 Advanced Web China

You can see that all items have appeared.

that's all

Recommended Posts