kaggle_tips.md

https://www.kaggle.com/arthurtok/introduction-to-ensembling-stacking-in-python

# apply

import pandas as pd
from sklearn.datasets import load_iris
iris = load_iris()
df = pd.DataFrame(iris.data, columns=df.feature_names)
df['HasBigSepal'] = df['sepal length (cm)'].apply(lambda x: 1 if x > 5 else 0)
df.head()
   sepal length (cm)  sepal width (cm)     ...       petal width (cm)  HasBigSepal
0                5.1               3.5     ...                    0.2            1
1                4.9               3.0     ...                    0.2            0
2                4.7               3.2     ...                    0.2            0
3                4.6               3.1     ...                    0.2            0
4                5.0               3.6     ...                    0.2            0

[5 rows x 5 columns]

# qcut

In [55]: pd.qcut(df['sepal length (cm)'], 5).head()
Out[55]:
0                   (5.0, 5.6]
1    (4.2989999999999995, 5.0]
2    (4.2989999999999995, 5.0]
3    (4.2989999999999995, 5.0]
4    (4.2989999999999995, 5.0]
Name: sepal length (cm), dtype: category
Categories (5, interval[float64]): [(4.2989999999999995, 5.0] < (5.0, 5.6] < (5.6, 6.1] < (6.1, 6.52] < (6.52, 7.9]]

In [56]: pd.qcut(df['sepal length (cm)'], 5)[0]
Out[56]: Interval(5.0, 5.6, closed='right')

In [57]: 1 in pd.qcut(df['sepal length (cm)'], 5)[0]
Out[57]: False

In [58]: 5.4 in pd.qcut(df['sepal length (cm)'], 5)[0]
Out[58]: True

In [65]: df['HasBigSepal'].head()
Out[65]:
0    1
1    0
2    0
3    0
4    0
Name: HasBigSepal, dtype: int64

In [66]: df['HasBigSepal'].map({0: 'small', 1: 'big'}).head()
Out[66]:
0      big
1    small
2    small
3    small
4    small
Name: HasBigSepal, dtype: object

分類分析

SGD

from sklearn.linear_model import SGDClassifier
clf = SGDClassifier()
clf.fit(X, y)
clf.score(X, y)

評価関数など