I prefer the solution using replace with a dict because of its simplicity and elegance:

df.replace({'-': None})

You can also have more replacements:

df.replace({'-': None, 'None': None})

And even for larger replacements, it is always obvious and clear what is replaced by what - which is way harder for long lists, in my opinion.

where is probably what you're looking for. So

data=data.where(data=='-', None) 

From the panda docs:

where [returns] an object of same shape as self and whose corresponding entries are from self where cond is True and otherwise are from other).

Before proceeding with this post, it is important to understand the difference between NaN and None. One is a float type, the other is an object type. Pandas is better suited to working with scalar types as many methods on these types can be vectorised. Pandas does try to handle None and NaN consistently, but NumPy cannot.

My suggestion (and Andy's) is to stick with NaN.

But to answer your question...

pandas >= 0.18: Use na_values=['-'] argument with read_csv

If you loaded this data from CSV/Excel, I have good news for you. You can quash this at the root during data loading instead of having to write a fix with code as a subsequent step.

Most of the pd.read_* functions (such as read_csv and read_excel) accept a na_values attribute.

file.csv

A,B
-,1
3,-
2,-
5,3
1,-2
-5,4
-1,-1
-,0
9,0

Now, to convert the - characters into NaNs, do,

import pandas as pd
df = pd.read_csv('file.csv', na_values=['-'])
df

     A    B
0  NaN  1.0
1  3.0  NaN
2  2.0  NaN
3  5.0  3.0
4  1.0 -2.0
5 -5.0  4.0
6 -1.0 -1.0
7  NaN  0.0
8  9.0  0.0

And similar for other functions/file formats.

P.S.: On v0.24+, you can preserve integer type even if your column has NaNs (yes, talk about having the cake and eating it too). You can specify dtype='Int32'

df = pd.read_csv('file.csv', na_values=['-'], dtype='Int32')
df

     A    B
0  NaN    1
1    3  NaN
2    2  NaN
3    5    3
4    1   -2
5   -5    4
6   -1   -1
7  NaN    0
8    9    0

df.dtypes

A    Int32
B    Int32
dtype: object

The dtype is not a conventional int type... but rather, a Nullable Integer Type. There are other options.

Handling Numeric Data: pd.to_numeric with errors='coerce

If you're dealing with numeric data, a faster solution is to use pd.to_numeric with the errors='coerce' argument, which coerces invalid values (values that cannot be cast to numeric) to NaN.

pd.to_numeric(df['A'], errors='coerce')

0    NaN
1    3.0
2    2.0
3    5.0
4    1.0
5   -5.0
6   -1.0
7    NaN
8    9.0
Name: A, dtype: float64

To retain (nullable) integer dtype, use

pd.to_numeric(df['A'], errors='coerce').astype('Int32')

0    NaN
1      3
2      2
3      5
4      1
5     -5
6     -1
7    NaN
8      9
Name: A, dtype: Int32 

To coerce multiple columns, use apply:

df[['A', 'B']].apply(pd.to_numeric, errors='coerce').astype('Int32')

     A    B
0  NaN    1
1    3  NaN
2    2  NaN
3    5    3
4    1   -2
5   -5    4
6   -1   -1
7  NaN    0
8    9    0

...and assign the result back after.

More information can be found in this answer.

With Pandas version ≥1.0.0, I would use DataFrame.replace or Series.replace:

df.replace(old_val, pd.NA, inplace=True)

This is better for two reasons:

1. It uses pd.NA instead of None or np.nan.
2. It optionally works in-place which could be more memory efficient depending upon the internal implementation.

df = pd.DataFrame(['-',3,2,5,1,-5,-1,'-',9])
df = df.where(df!='-', None)

Setting null values can be done with np.nan:

import numpy as np
df.replace('-', np.nan)

Advantage is that df.last_valid_index() recognizes these as invalid.

Using replace and assigning a new df:

import pandas as pd
df = pd.DataFrame(['-',3,2,5,1,-5,-1,'-',9])
dfnew = df.replace('-', 0)
print(dfnew)


(venv) D:\assets>py teste2.py
   0
0  0
1  3
2  2
3  5
4  1
5 -5

df.replace('-', np.nan).astype("object")

This will ensure that you can use isnull() later on your dataframe

Alternatively you can also use mask:

df.mask(df=='-', None)