Understanding ?weird? running behavior with @vaex.register_func() / min row wise

[yohplala]

Hi,

Trying to extract min row wise for a variable number of columns, I ended up with following implementations.

(@JovanVeljanoski , it works, at least in my test, do not try to find a bug. :) well I guess there isn't)

import vaex as vx
import numpy as np
import pandas as pd

@vx.register_function()
def min_row_wise(*args):
    print([col for col in args])
    return np.min(np.stack(args, axis=1), axis=1)

# Test data
df = pd.DataFrame({'val1': [0.5, 1, 6.3],
                   'val2': [0.8, 1.2, 3.2],
                   'val3': [1.5, 1.9, 4.0]})
vdf = vx.from_pandas(df)

# And let the magic be!
vdf.func.min_row_wise(*[vdf[col] for col in vdf.get_column_names()])

When doing so, with the print statement, one can see the calls to min_row_wise() with different arrays.
I find these calls surprising.

vdf.func.min_row_wise(*[vdf[col] for col in vdf.get_column_names()])
Out[45]: 
[array([0.5]), array([0.8]), array([1.5])]
[array([0.5]), array([0.8]), array([1.5])]
[array([0.5]), array([0.8]), array([1.5])]
[array([0.5, 1. , 6.3]), array([0.8, 1.2, 3.2]), array([1.5, 1.9, 4. ])]
[array([0.5]), array([0.8]), array([1.5])]
Expression = min_row_wise(val1, val2, val3)
Length: 3 dtype: float64 (expression)
-------------------------------------
0  0.5
1    1
2  3.2

min_row_wise() is called 3 times with only the 1st row of the DataFrame (1st value of each column).
Then with all values.
Then finally again with the 1st value.

@JovanVeljanoski, please, would it be possible to have some insights about this?

I thank you in advance for your feedback.
Have a good day,
Bests,

[maartenbreddels]

Nice work :)
Yes, to determine the dtype of the function we execute it with an array of length 1. I hope in the future we do this less often, so this is expected for now.

[yohplala]

Nice work :)

Thanks! Thank you to let users/devs discover about vaex actually! :)

Yes, to determine the dtype of the function we execute it with an array of length 1. I hope in the future we do this less often, so this is expected for now.

Ok, understood!

Thanks again!