CODEWITHSUNDEEP
pythonpandas
user308827
user308827

Reputation: 22021

Pandas merge dataframes based on closest match

I have the following 2 dataframes (df_a,df_b):

df_a

    N0_YLDF
0   11.79
1   7.86
2   5.78
3   5.35
4   6.32
5   11.79
6   6.89
7   10.74


df_b
    N0_YLDF N0_DWOC
0   6.29    4
1   2.32    4
2   9.10    4
3   4.89    4
4   10.22   4
5   3.80    3
6   5.55    3
7   6.36    3

I would like to add a column N0_DWOC in df_a, such that the value in that column is from the row where df_a['N0_YLDF'] is closest to df_b['N0_YLDF'].

Right now, I am doing a simple merge but that does not do what I want

Upvotes: 3

Views: 6584

Answers (4)

imak
imak

Reputation: 431

there is the exact method in pandas for that right now: pandas.merge_asof, REF: pandas documentation

In your case,

pd.merge_asof(df_a, df_b, on="N0_YLDF", direction="nearest")

Upvotes: 1

Dinesh Marimuthu
Dinesh Marimuthu

Reputation: 147

Another approach to this problem is to perform a Cartesian join followed by a absolute difference between the values of the common column

Then group by the column N0_YLDF to get the minimum value of the difference and use this again on the mergfed df to remerge again but this time using the merge as a filter. The explanation is insufficient but you might see what the code is doing.

mg = df_a.merge(df_b,how='cross')
mg['diff'] = mg.apply(lambda x:abs(x['N0_YLDF_x']-x['N0_YLDF_y']),axis=1 )
groups = mg.groupby('N0_YLDF_x')['diff'].min().reset_index()

mg.merge(groups.drop('N0_YLDF_x',axis=1), on='diff').drop(['N0_YLDF_y','diff'],axis=1).rename({'N0_YLDF_x':'N0_YLDF'})

output df

Upvotes: 0

JoeCondron
JoeCondron

Reputation: 8906

Another way is to do an subtract all pairs in the cartesian product and get the index of minimum absolute value for each one:

In [47]:ix = abs(np.atleast_2d(df_a['N0_YLDF']).T - df_b['N0_YLDF'].values).argmin(axis=1)
        ix
Out[47]: array([4, 2, 6, 6, 0, 4, 7, 4])

Then do 

df_a['N0_DWOC'] = df_b.ix[ix, 'N0_DWOC'].values

In [73]: df_a
Out[73]:
N0_YLDF N0_DWOC
0   11.79   4
1   7.86    4
2   5.78    3
3   5.35    3
4   6.32    4
5   11.79   4
6   6.89    3
7   10.74   4

Upvotes: 3

unutbu
unutbu

Reputation: 880547

You could find the cutoff values which are midway between the (sorted) values in df_b['N0_YLDF']. Then call pd.cut to categorize the values in df_a['N0_YLDF'], with the cutoff values being the bin edges:

import numpy as np
import pandas as pd

df_a = pd.DataFrame({ 'N0_YLDF': [11.79, 7.86, 5.78, 5.35, 6.32, 11.79, 6.89, 10.74]})
df_b = pd.DataFrame({ 'N0_YLDF':[6.29, 2.32, 9.10, 4.89, 10.22, 3.80, 5.55, 6.36] })

edges, labels = np.unique(df_b['N0_YLDF'], return_index=True)
edges = np.r_[-np.inf, edges + np.ediff1d(edges, to_end=np.inf)/2]
df_a['N0_DWOC'] = pd.cut(df_a['N0_YLDF'], bins=edges, labels=df_b.index[labels])
print(df_a)

yields

In [293]: df_a
Out[293]: 
   N0_YLDF N0_DWOC
0    11.79       4
1     7.86       2
2     5.78       6
3     5.35       6
4     6.32       0
5    11.79       4
6     6.89       7
7    10.74       4

To join the two DataFrames on N0_DWOC you could use:

print(df_a.join(df_b, on='N0_DWOC', rsuffix='_b'))

which yields

   N0_YLDF N0_DWOC  N0_YLDF_b
0    11.79       4      10.22
1     7.86       2       9.10
2     5.78       6       5.55
3     5.35       6       5.55
4     6.32       0       6.29
5    11.79       4      10.22
6     6.89       7       6.36
7    10.74       4      10.22

Upvotes: 4

Related Questions