I am still quite confused about the workings of multiindex in Pandas. I created a multi index as follows:
import pandas as pd
import numpy as np
arrays = [np.array(['pearson', 'pearson', 'pearson', 'pearson', 'spearman', 'spearman',
'spearman', 'spearman', 'kendall', 'kendall', 'kendall', 'kendall']),
np.array(['PROFESSIONAL', 'PROFESSIONAL', 'STUDENT', 'STUDENT',
'PROFESSIONAL', 'PROFESSIONAL', 'STUDENT', 'STUDENT',
'PROFESSIONAL', 'PROFESSIONAL', 'STUDENT', 'STUDENT']),
np.array(['r', 'p', 'r', 'p', 'rho', 'p', 'rho', 'p', 'tau', 'p', 'tau', 'p'])]
tuples = list(zip(*arrays))
index = pd.MultiIndex.from_tuples(tuples, names=['correlator', 'expertise', 'coeff-p'])
then I made an empty DataFrame out of them and added a column name 'pair':
results_df = pd.DataFrame(index=index)
results_df.columns.names = ['pair']
Filled with some toy data (results_df['attr1-attr2'] = [1,2,3,4,5,6,7,8,9,10,11,12]), it looks like this:
pair attr1-attr2
correlator expertise coeff-p
pearson PROFESSIONAL r 1
p 2
STUDENT r 3
p 4
spearman PROFESSIONAL rho 5
p 6
STUDENT rho 7
p 8
kendall PROFESSIONAL tau 9
p 10
STUDENT tau 11
p 12
However, instead of the dummy I want to add values from a dictionary. The dictionary has entries that look like this, for each attr-attr pair:
'attr-attr': {
'pearson': {
'STUDENT': {
'r': VALUE,
'p': VALUE
},
'PROFESSIONAL': {
'r': VALUE,
'p': VALUE
}
},
'spearman': {
'STUDENT': {
'r': VALUE,
'p': VALUE
},
'PROFESSIONAL': {
'r': VALUE,
'p': VALUE
}
}
'kendall': {
'STUDENT': {
'r': VALUE,
'p': VALUE
},
'PROFESSIONAL': {
'r': VALUE,
'p': VALUE
}
}
}
Actual example data below for you to work with:
correlations = {'NormNedit-NormEC_TOT': {'pearson': {'PROFESSIONAL': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}}, 'spearman': {'STUDENT': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}, 'PROFESSIONAL': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}, 'kendall': {'STUDENT': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}}, 'NormLiteral-NormEC_TOT': {'pearson': {'PROFESSIONAL': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}, 'STUDENT': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}, 'spearman': {'STUDENT': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}, 'PROFESSIONAL': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}}, 'kendall': {'STUDENT': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}}, 'NormHTra-NormEC_TOT': {'pearson': {'STUDENT': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}}, 'spearman': {'STUDENT': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}, 'PROFESSIONAL': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}}, 'kendall': {'STUDENT': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}}, 'NormScatter-NormEC_TOT': {'pearson': {'STUDENT': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}}, 'spearman': {'STUDENT': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}, 'PROFESSIONAL': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}, 'kendall': {'PROFESSIONAL': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}}, 'NormCrossS-NormEC_TOT': {'pearson': {'STUDENT': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}, 'PROFESSIONAL': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}}, 'spearman': {'STUDENT': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}, 'PROFESSIONAL': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}}, 'kendall': {'PROFESSIONAL': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}}, 'NormPdur-NormEC_TOT': {'pearson': {'STUDENT': {'r': 0.13615071018351657, 'p': 0.0002409555504769095}, 'PROFESSIONAL': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}}, 'spearman': {'STUDENT': {'rho': 0.10867061294616957, 'p': 0.003437711066527592}}, 'kendall': {'PROFESSIONAL': {'tau': 0.08185775947238913, 'p': 0.003435247172206748}}}}
So for each 'attr-attr' (topmost key) as column name, I want to add its values to the corresponding row in the multi index. However, I can't seem to find a way to do this in an efficient way. Missing values should be np.nan. I tried looping the dictionary and using query()[] but that didn't work.
for attr, attr_d in correlations.items():
for correl, correl_d in attr_d.items():
for split, split_d in correl_d.items():
results_df.query(f"correlator == {correl} and expertise == {split} and coeff_p == 'p'")[attr] = split_d['p']
results_df.query(f"correlator == {correl} and expertise == {split} and coeff_p != 'p'")[attr] = split_d['r'] if 'r' in split_d else split_d['rho'] if 'rho' in split_d else split['tau']
> pandas.core.computation.ops.UndefinedVariableError: name 'pearson' is not defined
I am aware that the data is relatively complex, so if something is not clear please let me know.
You can adapt Wouter Overmeire's answer to this question to make a multi-indexed dataframe out of your nested dictionary:
d = correlations
df = pd.DataFrame.from_dict({(i,j,k): d[i][j][k]
for i in d.keys()
for j in d[i].keys()
for k in d[i][j].keys()
}, orient='index').stack()
Then if you want the columns to come from the highest level of the nested dictionary (the attr-attr level), you can unstack the result:
df = df.unstack(level=0)
Note: I think there's a mistake in your sample data, where 'PROFESSIONAL': {'STUDENT': .... If that's not a mistake, and I'm just misunderstanding something, let me know.