pandas df.eval gives ValueError: data type must provide an itemsize

lisrael1 0
Viewed 29

I'll give a sort simple code that reproduces the issue that I have:

import pandas as pd

df = pd.DataFrame(dict(a=[1,2,3]))
df=df.eval('x=2')  # this one is ok
df.eval('y="num"')  # here it will fail

The error that I get is:

ValueError: data type must provide an itemsize

What is the issue? How can i make it work?
It wasnt like this at older pandas versions...

I know that I can replace it with:

df['y']="num"
# or
df.assign(y='num')

But this is not the answer that I need...

I also tried replacing "num" with:

np.str_("num")

Which do has .itemsize, but it didn't help...

Note that when using df.query, with another content, gives me that same issue which I'm trying to solve here. I'm just assuming it's the same issue.

eval valueerror pandas numpy
2 Answers

Use engine='python' parameter:

print(df.eval('y="num"', engine='python'))
   a  x    y
0  1  2  num
1  2  2  num
2  3  2  num
jezrael0

You should probably prefer assign here:

df.assign(y='num')

With multiple assignments and handling variable as strings

df.assign(**{'x': 2, 'y': 'num'})

output:

   a  x    y
0  1  2  num
1  2  2  num
2  3  2  num
mozway0
Related