Pandas：稀疏数据帧到没有 nan 值的字典

Question

我有一个大型稀疏数据框，sdf主要包含NaN在其中。当我使用sdf.to_dict()它时，它会输出该矩阵的密集版本，其中null填充了所有值。我如何省略这些NaN条目，并且只有输出条目对字典有价值？

例如，sdf是：

          2018-02-02  2018-02-03
23:58:36         NaN         NaN
23:58:37         1.0         NaN
23:58:40         NaN         NaN
23:58:41         NaN         NaN
23:58:42         NaN         NaN
23:58:43         NaN         NaN
23:58:48         NaN         NaN
23:58:49         NaN         NaN
23:58:50         NaN         NaN
23:58:52         NaN         1.0
23:58:59         NaN         NaN
23:59:00         NaN         NaN
23:59:01         NaN         NaN
23:59:05         NaN         NaN
23:59:07         NaN         NaN

stf.to_dict()会给出：

{'2018-02-02': {'23:58:36': nan, '23:58:37': 1.0, '23:58:40':
  nan, '23:58:41': nan, '23:58:42': nan, '23:58:43': nan,
  '23:58:48': nan, '23:58:49': nan, '23:58:50': nan, '23:58:52':
  nan, '23:58:59': nan, '23:59:00': nan, '23:59:01': nan,
  '23:59:05': nan, '23:59:07': nan}, '2018-02-03': {'23:58:36':
  nan, '23:58:37': nan, '23:58:40': nan, '23:58:41': nan,
  '23:58:42': nan, '23:58:43': nan, '23:58:48': nan, '23:58:49':
  nan, '23:58:50': nan, '23:58:52': 1.0, '23:58:59': nan,
  '23:59:00': nan, '23:59:01': nan, '23:59:05': nan, '23:59:07':
  nan}}

Evensdf是一个稀疏数据框。

---

抱歉含糊不清。我想保留所有非NaN条目。期望的输出是

{'2018-02-02': {'23:58:37': 1.0}, '2018-02-03': {'23:58:52': 1.0}}

Answer 1

stack与以下一起使用dict comprehension：

from collections import defaultdict
d = defaultdict(dict)
for (k1, k2), v in df.stack().items():
    d[k2][k1] = v

d1 = dict(d)

如果输入Series是DatetimeIndex：

print (s)
2018-02-02 23:58:37    1.0
2018-02-03 23:58:52    1.0
dtype: float64

from collections import defaultdict
d = defaultdict(dict)
for k, v in df.stack().items():
    d[k.strftime('%Y-%m-%d')][k.strftime('%H:%M:%S')] = v

d1 = dict(d)