[FIXED] Applying a function for only some indices

Issue

I have a function computeLeft which receives an index and returns four numbers. Something like this:

def computeLeft(i):
    return np.array([i*2, i*3, i*4, i*5])
    # edited to correct it

Right now in my code I use it like this:

import numpy as np
import pandas as pd

results=["val1","val2","val3","val4"]
df[results] = np.vectorize(computeLeft, signature="()->(4)")(range(len(df)))

where df is some dataframe.

This obviously applies the function to all rows of df. I want to apply this function to only some indexes of df.

So for example I have a list [2, 5, 7, 8, 10]. I want to compute computeLeft only for the indices in the list and that the columns in result have values only for those rows (the rest having Nan).

How can I apply computeLeft selectively like this?

Solution

First you need to fix your function and its vectorization:

def computeLeft(i):
    return np.array([i*2, i*3, i*4, i*5])

computeLeftVectorized = np.vectorize(computeLeft, signature="()->(n)")

However, as mentioned by @jared in his anwser, it is much faster to just write the function as a pure numpy function in the first case.

And to achieve your goal you can use loc[]:

results=["val1","val2","val3","val4"]

# Add new columns and fill the columns with NaN
df[results] = np.nan

indices_to_change = [2,5,7,8,10]
# change all rows in results with the provided indices
df.loc[indices_to_change, results] = computeLeftVectorized(indices_to_change)

This assumes that your dataframe has index labels that are the given integers (the given indices are not just the row number with different index labels).

Answered By - Oskar Hofmann