Usage Examples

Here’s a small python snippet that estimates the polarization (nDFU) values for three distinct annotation distributions:

import numpy as np
from apunim import dfu

rng = np.random.default_rng(42)
data = rng.normal(loc=0, scale=1, size=1000)
score = dfu(data, bins=10, normalized=True)
print(f"Normal distribution polarization: {score:.4f}")


mode1 = rng.normal(loc=-2, scale=0.3, size=500)
mode2 = rng.normal(loc=2, scale=0.3, size=500)
data = np.hstack([mode1, mode2])
score = dfu(data, bins=10, normalized=True)
print(f"Bimodal distribution polarization: {score:.4f}")

# Output:
Normal distribution polarization: 0.0000
Bimodal distribution polarization: 0.6024

And a snippet that investigates whether the polarization of a discussion of two comments is caused by gender:

import numpy as np
from apunim import aposteriori_unimodality

# Example discussion with two comments (c1, c2),
# each annotated by two gender groups (A, B).

# c1: polarized (A gives low scores, B gives high)
c1_A = [1, 1, 2]
c1_B = [5, 5, 4]

# c2: also polarized but differently distributed
c2_A = [2, 2, 3]
c2_B = [4, 5, 5]

annotations = c1_A + c1_B + c2_A + c2_B
factor_group = ["A"] * len(c1_A) + ["B"] * len(c1_B) + \
               ["A"] * len(c2_A) + ["B"] * len(c2_B)
comment_group = ["c1"] * (len(c1_A) + len(c1_B)) + \
                  ["c2"] * (len(c2_A) + len(c2_B))

result = aposteriori_unimodality(
      annotations,
      factor_group,
      comment_group,
      num_bins=5,
)

for group, res in result.items():
      print(f"Group: {group}")
      print(f"  Apunim: {res.apunim:.4f}")
      print(f"  p-value: {res.pvalue:.4f}")

# Output:
Group: A
   Apunim: -1.3529
   p-value: 0.1114
Group: B
   Apunim: -1.7778
   p-value: 0.1252

# Output:
Normal distribution polarization: 0.0000
Bimodal distribution polarization: 0.6024