Bizarre behaviour of pandas Series.value_counts()

问题

I have a Pandas Series with numerical data and I want to find its unique values together with their frequency-appearance. I use the standard procedure

# Given the my_data is a column of a pd.Dataframe df
unique = df[my_data].value_counts() 
print unique

And here is the results that I get

# -------------------OUTPUT
-0.010000    46483 
-0.010000    16895
-0.027497    12215
-0.294492    11915
 0.027497    11397

What I don't get is why I have the "same value" (-0.01) occurring twice. Is that an internal threshold (small value) or is something that I am doing wrong??

Update

If I store the dataframe in csv and read it again I get the correct result, namely:

# -------------------OUTPUT -0.010000 63378 -0.027497 12215 -0.294492 11915 0.027497 11397 Solution

Based on the discussion, I found the source of the problem and the solution. As mentioned it is a floating-point precision which can be solved with rounding the values. Though, I wouldn't be able to see that without

pd.set_option('display.float_format', repr)

Thanks a lot for the help!!

回答1:

I think it's a float precision issue similar to the following one:

In [1]: 0.1 + 0.2
Out[1]: 0.30000000000000004

In [2]: 0.1 + 0.2 == 0.3
Out[2]: False

so try this:

df[my_data].round(6).value_counts() 

---

UPDATE:

Demo:

In [14]: s = pd.Series([-0.01, -0.01, -0.01000000000123, 0.2])

In [15]: s
Out[15]:
0   -0.01
1   -0.01
2   -0.01
3    0.20
dtype: float64

In [16]: s.value_counts()
Out[16]:
-0.01    2
-0.01    1
 0.20    1
dtype: int64

In [17]: s.round(6).value_counts()
Out[17]:
-0.01    3
 0.20    1
dtype: int64

来源：https://stackoverflow.com/questions/50105629/bizarre-behaviour-of-pandas-series-value-counts