熊猫数据帧/Numpy 数组“轴”定义的模糊性

对于如何定义 python 轴，以及它们是引用 DataFrame 的行还是列，我感到非常困惑。考虑下面的代码:

>>> df = pd.DataFrame([[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3]], columns=["col1", "col2", "col3", "col4"])
>>> df
col1  col2  col3  col4
0     1     1     1     1
1     2     2     2     2
2     3     3     3     3

因此，如果我们调用 df.mean(axis=1)，我们将得到横跨行的平均值:

>>> df.mean(axis=1)
0    1
1    2
2    3

然而，如果我们调用 df.drop(name, axis=1)，我们实际上是 放下一根柱子，而不是一行:

>>> df.drop("col4", axis=1)
col1  col2  col3
0     1     1     1
1     2     2     2
2     3     3     3

谁能帮我理解一下熊猫/numpy/scypy 中的“轴”是什么意思？

值得注意的是，DataFrame.mean的定义可能是错误的。在 DataFrame.mean的文档中说，axis=1应该表示列上的平均值，而不是行上的平均值。

31613 次浏览
小开
最佳答案

It's perhaps simplest to remember it as 0=down and 1=across.

This means:

  • Use axis=0 to apply a method down each column, or to the row labels (the index).
  • Use axis=1 to apply a method across each row, or to the column labels.

Here's a picture to show the parts of a DataFrame that each axis refers to:

It's also useful to remember that Pandas follows NumPy's use of the word axis. The usage is explained in NumPy's glossary of terms:

Axes are defined for arrays with more than one dimension. A 2-dimensional array has two corresponding axes: the first running vertically downwards across rows (axis 0), and the second running horizontally across columns (axis 1). [my emphasis]

So, concerning the method in the question, df.mean(axis=1), seems to be correctly defined. It takes the mean of entries horizontally across columns, that is, along each individual row. On the other hand, df.mean(axis=0) would be an operation acting vertically downwards across rows.

Similarly, df.drop(name, axis=1) refers to an action on column labels, because they intuitively go across the horizontal axis. Specifying axis=0 would make the method act on rows instead.

小开

Another way to explain:

// Not realistic but ideal for understanding the axis parameter
df = pd.DataFrame([[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3]],
columns=["idx1", "idx2", "idx3", "idx4"],
index=["idx1", "idx2", "idx3"]
)


---------------------------------------1
|          idx1  idx2  idx3  idx4
|    idx1     1     1     1     1
|    idx2     2     2     2     2
|    idx3     3     3     3     3
0

About df.drop (axis means the position)

A: I wanna remove idx3.
B: **Which one**? // typing while waiting response: df.drop("idx3",
A: The one which is on axis 1
B: OK then it is >> df.drop("idx3", axis=1)


// Result
---------------------------------------1
|          idx1  idx2     idx4
|    idx1     1     1     1
|    idx2     2     2     2
|    idx3     3     3     3
0

About df.apply (axis means direction)

A: I wanna apply sum.
B: Which direction? // typing while waiting response: df.apply(lambda x: x.sum(),
A: The one which is on *parallel to axis 0*
B: OK then it is >> df.apply(lambda x: x.sum(), axis=0)


// Result
idx1    6
idx2    6
idx3    6
idx4    6
小开

There are already proper answers, but I give you another example with > 2 dimensions.

The parameter axis means axis to be changed.
For example, consider that there is a dataframe with dimension a x b x c.

  • df.mean(axis=1) returns a dataframe with dimenstion a x 1 x c.
  • df.drop("col4", axis=1) returns a dataframe with dimension a x (b-1) x c.

Here, axis=1 means the second axis which is b, so b value will be changed in these examples.

小开

It should be more widely known that the string aliases 'index' and 'columns' can be used in place of the integers 0/1. The aliases are much more explicit and help me remember how the calculations take place. Another alias for 'index' is 'rows'.

When axis='index' is used, then the calculations happen down the columns, which is confusing. But, I remember it as getting a result that is the same size as another row.

Let's get some data on the screen to see what I am talking about:

df = pd.DataFrame(np.random.rand(10, 4), columns=list('abcd'))
a         b         c         d
0  0.990730  0.567822  0.318174  0.122410
1  0.144962  0.718574  0.580569  0.582278
2  0.477151  0.907692  0.186276  0.342724
3  0.561043  0.122771  0.206819  0.904330
4  0.427413  0.186807  0.870504  0.878632
5  0.795392  0.658958  0.666026  0.262191
6  0.831404  0.011082  0.299811  0.906880
7  0.749729  0.564900  0.181627  0.211961
8  0.528308  0.394107  0.734904  0.961356
9  0.120508  0.656848  0.055749  0.290897

When we want to take the mean of all the columns, we use axis='index' to get the following:

df.mean(axis='index')
a    0.562664
b    0.478956
c    0.410046
d    0.546366
dtype: float64

The same result would be gotten by:

df.mean() # default is axis=0
df.mean(axis=0)
df.mean(axis='rows')

To get use an operation left to right on the rows, use axis='columns'. I remember it by thinking that an additional column may be added to my DataFrame:

df.mean(axis='columns')
0    0.499784
1    0.506596
2    0.478461
3    0.448741
4    0.590839
5    0.595642
6    0.512294
7    0.427054
8    0.654669
9    0.281000
dtype: float64

The same result would be gotten by:

df.mean(axis=1)

Add a new row with axis=0/index/rows

Let's use these results to add additional rows or columns to complete the explanation. So, whenever using axis = 0/index/rows, its like getting a new row of the DataFrame. Let's add a row:

df.append(df.mean(axis='rows'), ignore_index=True)


a         b         c         d
0   0.990730  0.567822  0.318174  0.122410
1   0.144962  0.718574  0.580569  0.582278
2   0.477151  0.907692  0.186276  0.342724
3   0.561043  0.122771  0.206819  0.904330
4   0.427413  0.186807  0.870504  0.878632
5   0.795392  0.658958  0.666026  0.262191
6   0.831404  0.011082  0.299811  0.906880
7   0.749729  0.564900  0.181627  0.211961
8   0.528308  0.394107  0.734904  0.961356
9   0.120508  0.656848  0.055749  0.290897
10  0.562664  0.478956  0.410046  0.546366

Add a new column with axis=1/columns

Similarly, when axis=1/columns it will create data that can be easily made into its own column:

df.assign(e=df.mean(axis='columns'))


a         b         c         d         e
0  0.990730  0.567822  0.318174  0.122410  0.499784
1  0.144962  0.718574  0.580569  0.582278  0.506596
2  0.477151  0.907692  0.186276  0.342724  0.478461
3  0.561043  0.122771  0.206819  0.904330  0.448741
4  0.427413  0.186807  0.870504  0.878632  0.590839
5  0.795392  0.658958  0.666026  0.262191  0.595642
6  0.831404  0.011082  0.299811  0.906880  0.512294
7  0.749729  0.564900  0.181627  0.211961  0.427054
8  0.528308  0.394107  0.734904  0.961356  0.654669
9  0.120508  0.656848  0.055749  0.290897  0.281000

It appears that you can see all the aliases with the following private variables:

df._AXIS_ALIASES
{'rows': 0}


df._AXIS_NUMBERS
{'columns': 1, 'index': 0}


df._AXIS_NAMES
{0: 'index', 1: 'columns'}
小开

When axis='rows' or axis=0, it means access elements in the direction of the rows, up to down. If applying sum along axis=0, it will give us totals of each column.

When axis='columns' or axis=1, it means access elements in the direction of the columns, left to right. If applying sum along axis=1, we will get totals of each row.

Still confusing! But the above makes it a bit easier for me.

小开

I remembered by the change of dimension, if axis=0, row changes, column unchanged, and if axis=1, column changes, row unchanged.


提交答案