如何在熊貓中將多列折疊為一列

Question

我有一個填充了用戶和類別的 Pandas 數據框，但這些類別有多個列。

|   user  |       category    | val1 | val2 | val3 |
| ------  | ------------------| -----| ---- | ---- |
| user 1  | c1                |   3  |  NA  | None |
| user 1  | c2                |   NA |  4   | None |
| user 1  | c3                |   NA |  NA  | 7    |
| user 2  | c1                |   5  |  NA  | None |
| user 2  | c2                |   NA |  7   | None |
| user 2  | c3                |   NA |  NA  | 2    |

我想得到它，以便將值壓縮到單個列中。

|   user  |       category    | value|
| ------  | ------------------| -----| 
| user 1  | c1                |   3  | 
| user 1  | c2                |   4  | 
| user 1  | c3                |   7  |
| user 2  | c1                |   5  | 
| user 2  | c2                |   7  | 
| user 2  | c3                |   2  |

最終，得到如下矩陣：

np.array([[3, 4, 7], [5, 7, 2]])

Answer 1

您可以使用pd.DataFrame.bfill回填所選列的值。

val_cols = ['val1', 'val2', 'val3']

df['value'] = pd.to_numeric(df[val_cols].bfill(axis=1).iloc[:, 0], errors='coerce')

print(df)

    user0 category  val1  val2  val3  value
0  user 1       c1   3.0   NaN  None    3.0
1  user 1       c2   NaN   4.0  None    4.0
2  user 1       c3   NaN   NaN  7       7.0
3  user 2       c1   5.0   NaN  None    5.0
4  user 2       c2   NaN   7.0  2       7.0
5  user 2       c3   NaN   NaN  2       2.0

Answer 2

• 首先使用['user', 'category']設置索引
• 接着查找第一個非空值的位置
• 使用您設置的相同索引和您查找的新值構建一個熊貓系列

---

d = df.set_index(['user', 'category'])
pd.Series(d.lookup(d.index, d.isna().idxmin(1)), d.index).reset_index(name='value')

     user category value
0  user 1       c1     3
1  user 1       c2     4
2  user 1       c3     7
3  user 2       c1     5
4  user 2       c2     7
5  user 2       c3     2

您可以跳過索引的重置並取消堆疊以獲得最終結果

d = df.set_index(['user', 'category'])
pd.Series(d.lookup(d.index, d.isna().idxmin(1)), d.index).unstack()

category c1 c2 c3
user             
user 1    3  4  7
user 2    5  7  2

Answer 3

您可以簡單地fillna(0) ( df2 = df.fillna(0) ) 並使用| 操作員。

先轉換成int

df2.loc[:, ['val1','val2','val3']] = df2[['val1','val2','val3']].astype(int)

然后

df2['val4'] = df2.val1.values | df2.val2.values | df2.val3.values