Question

我有一个pandas数据框：

>>df_freq = pd.DataFrame([["Z11", "Z11", "X11"], ["Y11","",""], ["Z11","Z11",""]], columns=list('ABC'))

>>df_freq
    A   B   C
0   Z11 Z11 X11
1   Y11     
2   Z11 Z11

我想确保每一行只有唯一值。因此它应该是这样的：删除的值可以替换为零或空

    A   B   C
0   Z11 0   X11
1   Y11     
2   Z11 0

我的数据框很大，有数百列和数千行。目标是计算该数据框中的唯一值。我通过将数据帧转换为矩阵并应用

来实现这一点

>>np.unique(mat.astype(str), return_counts=True)

但是在某些行中会出现相同的值，我想在应用np.unique（）方法之前将其删除。我想在每一行中保留唯一值。

Answer 1

使用astype(bool)和duplicated

的组合

mask = df_freq.apply(pd.Series.duplicated, 1) & df_freq.astype(bool)

df_freq.mask(mask, 0)

     A  B    C
0  Z11  0  X11
1  Y11        
2  Z11  0

Answer 2

这是一个矢量化的NumPy方法 -

def reset_rowwise_dups(df):
    n = df.shape[0]
    row_idx = np.arange(n)[:,None]

    a = df_freq.values
    idx = np.argsort(a,1)
    sorted_a = a[row_idx, idx]
    idx_reversed = idx.argsort(1)
    sorted_a_dupmask = sorted_a[:,1:] == sorted_a[:,:-1]
    dup_mask = np.column_stack((np.zeros(n,dtype=bool), sorted_a_dupmask))
    final_mask = dup_mask[row_idx, idx_reversed] & (a != '' )
    a[final_mask] = 0

示例运行 -

In [80]: df_freq
Out[80]: 
     A    B    C    D
0  Z11  Z11  X11  Z11
1  Y11            Y11
2  Z11  Z11       X11

In [81]: reset_rowwise_dups(df_freq)

In [82]: df_freq
Out[82]: 
     A  B    C    D
0  Z11  0  X11    0
1  Y11            0
2  Z11  0       X11

运行时测试

# Proposed earlier in this post
def reset_rowwise_dups(df):
    n = df.shape[0]
    row_idx = np.arange(n)[:,None]

    a = df.values
    idx = np.argsort(a,1)
    sorted_a = a[row_idx, idx]
    idx_reversed = idx.argsort(1)
    sorted_a_dupmask = sorted_a[:,1:] == sorted_a[:,:-1]
    dup_mask = np.column_stack((np.zeros(n,dtype=bool), sorted_a_dupmask))
    final_mask = dup_mask[row_idx, idx_reversed] & (a != '' )
    a[final_mask] = 0

# @piRSquared's soln using pandas apply
def apply_based(df):
    mask = df.apply(pd.Series.duplicated, 1) & df.astype(bool)
    return df.mask(mask, 0)

计时 -

In [151]: df_freq = pd.DataFrame([["Z11", "Z11", "X11", "Z11"], \
     ...:  ["Y11","","", "Y11"],["Z11","Z11","","X11"]], columns=list('ABCD'))

In [152]: df_freq
Out[152]: 
     A    B    C    D
0  Z11  Z11  X11  Z11
1  Y11            Y11
2  Z11  Z11       X11

In [153]: df = pd.concat([df_freq]*10000,axis=0)

In [154]: df.index = range(df.shape[0])

In [155]: %timeit apply_based(df)
1 loops, best of 3: 3.35 s per loop

In [156]: %timeit reset_rowwise_dups(df)
100 loops, best of 3: 12.7 ms per loop

Answer 3

def replaceDuplicateData(nestedList):
    for row in range(len(nestedList)):
        uniqueDataRow = []
        for col in range(len(nestedList[row])):
            if nestedList[row][col] not in uniqueDataRow:
                uniqueDataRow.append(nestedList[row][col])
            else:
                nestedList[row][col] = 0
    return nestedList

nestedList = [["Z11", "Z11", "X11"], ["Y11","",""], ["Z11","Z11",""]]
print (replaceDuplicateData(nestedList))

基本上，您可以使用上述功能删除矩阵中的重复项。

删除一行pandas数据帧中的重复值

3 个答案: