0

I have a DataFrame like so:

import pandas as pd

d = {'param_1': [1.0, 2.0, 1.0, 1.0, 3.0, 3.0, 1.0, 2.0, 2.0,], 
     'param_2': [0.02, 0.08, 0.02, 0.08, 0.08, 0.02, 0.02, 0.08, 0.02],
     'param_3': [0.8, 1.6, 2.5, 1.6, 1.6, 2.5, 0.8, 1.6, 0.8],
     'param_4': [0, 1, 0, 0, 1, 0, 0, 1, 0],
     'output_value': [67.85, 32.56, 126.4, -56.32, -10.36, 3.67, -6.78, -96.8, 325.4]
    }

df = pd.DataFrame(d)
print(df)

      param_1  param_2  param_3  param_4  output_value
0      1.0     0.02      0.8        0          67.85
1      2.0     0.08      1.6        1          32.56
2      1.0     0.02      2.5        0         126.40
3      1.0     0.08      1.6        0         -56.32
4      3.0     0.08      1.6        1         -10.36
5      3.0     0.02      2.5        0           3.67
6      1.0     0.02      0.8        0          -6.78
7      2.0     0.08      1.6        1         -96.80
8      2.0     0.02      0.8        0         325.40

I would like to identify the combinations of params (in the above example, rows 1 and 7 are the same combination of params, which resulted in different output_values) , then sum the output_value from each row where that combination of parameters appears. I would also like to count the number of times that combination of parameters appears.

I came up with this: 

for col in df.columns[0:4]:
    for each in df[col].unique():
        print(col, ":", each, "value sum:", df[df[col] == each]['output_value'].sum(), "... number of unique values:", pd.Series(df[df[col] == each]['output_value'].unique()).count())

It’s useful, but wrong: the sums include duplicate values, and the params are all individualized.

Output should be something like this, for each combination of parameters that appears, or param_set:

param_set: {param_1: 2.0, param_2: 0.08, param_3: 1.6, param_4: 1} 
output_value_sum = -64.24
n_rows_for_this_set = 2

param_set: {param_1: 1.0, param_2: 0.02, param_3: 0.8, param_4: 0} 
output_value_sum = -61.07
n_rows_for_this_set = 2

This similar question: unique combinations of values in selected columns in pandas data frame and count seems to answer the count question, and this: df.groupby(list(df.columns[0:4])).size() returns part of what I need.

tw0000
  • 475
  • 1
  • 7
  • 13

1 Answers1

0

Try using df.groupby(['param_1', 'param_2', 'param_3', 'param_4']).agg(['sum', 'count']), which returns:

                                 output_value      
                                         sum count
param_1 param_2 param_3 param_4                   
1.0     0.02    0.8     0              61.07     2
                2.5     0             126.40     1
        0.08    1.6     0             -56.32     1
2.0     0.02    0.8     0             325.40     1
        0.08    1.6     1             -64.24     2
3.0     0.02    2.5     0               3.67     1
        0.08    1.6     1             -10.36     1

You could unpack this DataFrame to print it out the way you'd prefer to see it, or just view it this way!

tw0000
  • 475
  • 1
  • 7
  • 13
  • Did you answer you're own question or are you satisfied with your solution? – Scott Boston Jul 10 '18 at 12:33
  • I did answer my own question @ScottBoston, not sure if I did this in the correct manner, although StackOverflow seemed to invite me to with their "Answer your own question!" check box :) – tw0000 Jul 10 '18 at 21:24