Python熊猫read_csv：我们可以在一行中将STRING加载到NUMPY吗？

Question

I have confused with parameters of read_csv of panda.

I wanna make a classifier with Support Vector Machines. To use classifier I need both vectors X and Y to be numpy. I got a csv file in which there are TWO columns:

1. the first column is a number(target), for instance 1 or 0

2. the second column is a vector(feature) with " " seperator, for instance 12 32 63 73 563 34.

The problem I ran into:

1. values from the first column are being loaded as 'numpy.int32'

2. values from the second column are being loaded as 'str' while I want them to be numpy arrays.

    import pandas as pd import numpy as np DF = pd.read_csv("C:\\\\STUFF\\\\foo.csv") df = DF.head(2) X = df["firstcol"] target = X.values for i in target: print (type(i)) Y = df["secondcol"] feature = Y.values for j in feature: print (type(j)) 

So the output is

<class 'numpy.int64'>
<class 'numpy.int64'>
<class 'str'>
<class 'str'>

The question is: What is the fastest and appropriate way to transform second column into numpy?

Answer 1

尝试这个：

df["secondcol"].apply(lambda x: np.array(x.split()).astype(int))

Answer 2

This works for me:

t = ['12 32 63 73 563 34']
y = [int(x) for x in str.split(" ") for str in t]
print(y)

prints: [12, 32, 63, 73, 563, 34] . This only works if all the cells are in the format you specified and there are not any letters in there.

Answer 3

Supposing your csv file look like that :

1,12 32 63 73 563 34
2,12 32 63 73 563 33
4,12 32 63 73 563 35

the more logic way to read it is :

df=pd.read_csv('data.csv',header=None,sep='[ ,]',engine='python',index_col=0)

then you have directly you data in cols, with first column as index. each row is (like) a numpy array.

In [4]: df
Out[4]: 
    1   2   3   4    5   6
0                         
1  12  32  63  73  563  34
2  12  32  63  73  563  33
4  12  32  63  73  563  35

In [5]: df.loc[4]
Out[5]: 
1     12
2     32
3     63
4     73
5    563
6     35
Name: 4, dtype: int64

In [6]: df.loc[4].values
Out[6]: array([ 12,  32,  63,  73, 563,  35], dtype=int64)


In [7]: df.loc[4].sum()
Out[7]: 778