Plugin "marginal_distributions" does not sample from marginal distributions

[gg-tmp]

Description

I had the impression that the plugin "marginal_distributions" is supposed to sample (independently) from the marginal distributions of the given dataset. But, in fact, it does not seem to sample at all!

How to Reproduce

A brief experiment demonstrates the problem.

from synthcity.plugins import Plugins
from synthcity.plugins.core.dataloader import GenericDataLoader
import pandas as pd
import numpy as np

# Generate sample of two independent Variables

# Parameters
N = 1_000  # Sample size
p = 0.6  # Probability P(X1 = "I")
q = 0.3  # Probability P(X2 = "A")

# Categoricals
v1 = ["I", "II"]
X1 = np.random.choice(v1, size=N, p=[p, 1 - p])
v2 = ["A", "B"]
X2 = np.random.choice(v2, size=N, p=[q, 1 - q])

# # Integers
# v1 = [-2, -1]
# X1 = np.random.choice(v1, size=N, p=[p, 1 - p])
# v2 = [ 1, 2]
# X2 = np.random.choice(v2, size=N, p=[q, 1 - q])

data = GenericDataLoader(list(zip(X1, X2)))
print(data)

syn_model = Plugins().get("marginal_distributions")

syn_model.fit(data)
tt = syn_model.generate(N).dataframe()
print(syn_model.sampling_strategy)

print(tt)

df = pd.DataFrame({
    '0': v1,
    '1': v2
}, dtype=tt["0"].dtype)
# Replicate the DataFrame N/2 times
df_replicated = pd.concat([df] * int(N / 2), ignore_index=True)

print(tt.equals(df_replicated))

Expected Behavior

I would have expected that

1. tt is random
2. the marginal distributions of tt are (up to sampling error) equal to the marginal distributions of data, i.e. p respectively q.

System Information

• OS: Microsoft Windows [Version 10.0.22631.4317]
• Language Version: Python 3.11.9
• synthcity Version: 0.2.11