所以我有这个 pandas.Dataframe
C1 C2 C3 C4 C5 Start End C8
A 1 - - - 1 4 -
A 2 - - - 6 10 -
A 3 - - - 11 14 -
A 4 - - - 15 19 -
其中 - 是对象,Start 是初始坐标,end 是每个元素的最终坐标。
我定义了这个函数来计算表中所有区间的并集,在这个例子中应该总和为 [1,19]-{5}(基本上是一个包含所有包含元素的 numpy 数组)。
def coverage(table):
#return a dataframe with the coverage of each individual peptide in a protein
interval = (table.apply(lambda row : range(int(row['Start']),int(row['End'])+1),axis=1))]
#if there is only one peptide, return the range between its start and end positions
if len(table) == 1: return asarray(range(int(table['Start']),int(table['End'])+1))
#if there are more, unite all the intervals
if len(table) > 1:
return reduce(union1d,(list(interval)))
所以我将该函数迭代地应用于多个 DataFrame(第一个是 A,然后是 B、C 等)。问题是对于某些表,这会失败并给出此错误:
Traceback (most recent call last):
File "At_coverage.py", line 37, in <module>
covdir[prot] = coverage(data)
File "At_coverage.py", line 21, in coverage
interval = (table.apply(lambda row : range(int(row['Start']),int(row['End'])+1),axis=1))
File "/usr/lib/python2.7/dist-packages/pandas/core/frame.py", line 3312, in apply
return self._apply_standard(f, axis, reduce=reduce)
File "/usr/lib/python2.7/dist-packages/pandas/core/frame.py", line 3417, in _apply_standard
result = self._constructor(data=results, index=index)
File "/usr/lib/python2.7/dist-packages/pandas/core/frame.py", line 201, in __init__
mgr = self._init_dict(data, index, columns, dtype=dtype)
File "/usr/lib/python2.7/dist-packages/pandas/core/frame.py", line 323, in _init_dict
dtype=dtype)
File "/usr/lib/python2.7/dist-packages/pandas/core/frame.py", line 4473, in _arrays_to_mgr
return create_block_manager_from_arrays(arrays, arr_names, axes)
File "/usr/lib/python2.7/dist-packages/pandas/core/internals.py", line 3760, in create_block_manager_from_arrays
construction_error(len(arrays), arrays[0].shape[1:], axes, e)
File "/usr/lib/python2.7/dist-packages/pandas/core/internals.py", line 3732, in construction_error
passed,implied))
ValueError: Shape of passed values is (7,), indices imply (7, 8)
失败的 DataFrame 如下:
Protein Peptide \
11106 sp|Q75W54|EBM_ARATH GJDGFJK
11107 sp|Q75W54|EBM_ARATH GJDGFJK
11108 sp|Q75W54|EBM_ARATH JJDPHJVSTFFDDYKR
11109 sp|Q75W54|EBM_ARATH JJDPHJVSTFFDDYKR
11110 sp|Q75W54|EBM_ARATH JNGEPJFJR
11111 sp|Q75W54|EBM_ARATH JNGEPJFJR
11112 sp|Q75W54|EBM_ARATH JNGEPJFJR
Fraction Count \
11106 AT_indark_IEX_fraction_18a_20150422.uniprot-pr... 2
11107 AT_indark_IEX_fraction_21a_20150422.uniprot-pr... 2
11108 AT_indark_IEX_fraction_18a_20150422.uniprot-pr... 2
11109 AT_indark_IEX_fraction_19a_20150422.uniprot-pr... 1
11110 AT_indark_IEX_fraction_19a_20150422.uniprot-pr... 2
11111 AT_indark_IEX_fraction_22a_20150422.uniprot-pr... 2
11112 AT_indark_IEX_fraction_25a_20150422.uniprot-pr... 2
Sequence Start End Length
11106 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 577 584 944
11107 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 577 584 944
11108 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 210 226 944
11109 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 210 226 944
11110 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 344 353 944
11111 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 344 353 944
11112 MAEIGKTVLDFGWIAARSTEVDVNGVQLTTTNPPAISSESRWMEAA... 344 353 944
[7 rows x 8 columns]
为了让它工作,我将第三行替换为
interval = (table.apply(lambda row : range(int(row['Start']),int(row['End'])+4),axis=1)).apply(lambda row: row[:-3])
我注意到它也适用于除 +1 以外的任何其他数字(尽管对于其他一些数字,它会在循环稍后的另一个 DataFrame 处崩溃。
所以这个解决方案是多余的和愚蠢的。我的假设是这个特定数据框中的行数与一些奇怪的参数(比如列数或类似的东西)相匹配,这使得 Pandas 试图简化某些东西然后它崩溃了。
我制作了一个简化版本的程序,它也适用于多个起点和终点:
def multicov(row):
intervals = []
for i in range(len(row['Start'])):
#print data
intervals.append((range(int(row['Start'][i]),int(row['End'][i])+1)))
return reduce(union1d,intervals)
dir = {'Start':[[1,7],[14]],
'End':[[5,10],[18]]}
df = DataFrame(dir,columns=['Start','End'])
print df
print df.apply(multicov,axis=1)
在这种情况下,它给出了同样的错误
ValueError: Shape of passed values is (2,), indices imply (2, 2)
但有趣的是,如果我从函数返回两个元素(以便它匹配 2,2)它表现良好。
return reduce(union1d,intervals),'foobar'
Start End
0 [1, 7] [5, 10]
1 [14] [18]
[2 rows x 2 columns]
0 ([1, 2, 3, 4, 7, 8, 9, 10], foobar)
1 ([14, 15, 16, 17, 18], foobar)
dtype: object
如果我将输出指定为列表,
return [reduce(union1d,intervals),'foobar']
它将前面的列名与输出相匹配!
Start End
0 [1, 7] [5, 10]
1 [14] [18]
[2 rows x 2 columns]
Start End
0 [1, 2, 3, 4, 7, 8, 9] foobar
1 [14, 15, 16, 17] foobar
[2 rows x 2 columns]
所以我认为该错误与 Pandas 试图强制我以前的 DataFrame 和输出的 DataFrame 之间存在某种兼容性有关,但令我惊讶的是,对于大多数 DataFrame,它运行良好!
最佳答案
方法 apply(func)
遍历行(或列)并将 func
应用于每一行。 func
的结果然后被放入一个新的数据框或一个系列中。如果 func
返回一个标量值(例如 sum
所做的),那么它就是一个系列。如果它返回一个数组、列表或系列,那么结果是一个维度框架,具体取决于该数组的长度。
在您的代码中,func
返回不同长度(间隔长度)的数组,这些数组不能放在一个框架中。因此错误。 (实际上,您遇到的第一个错误可能是这样的:ValueError: could not broadcast input array from shape (5) into shape (9)
。)
线
return reduce(union1d,intervals),'foobar'
返回一个元组,所以 apply
的结果是一个系列。和
return [reduce(union1d,intervals),'foobar']
返回一个长度为 2 的列表。因此,您在这里得到一个 n x 2
数据框。维度与输入数据框一致,因此 pandas 假定您想要修改原始框架的单元格(类似于应用 lambda x: 2*x
)并且它保留列名。
一个可能有效的解决方案是将函数中的 range(x, y)
更改为 tuple(range(x, y))
。但它既不高效也不 pythonic。更好的方法是用行上的显式循环替换 apply
,例如:
def coverage(table):
intervals = []
for row in table.itertuples():
intervals += list(range(row.Start, row.End + 1))
return np.unique(intervals)
关于python - pandas.apply 期望输出形状(传递值的形状是(x,),索引暗示(x,y)),我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/38274014/