data trove witch trials

This column represents a historical year, likely a creation, publication, or event date for records spanning 1300 to 1850 — consistent with a cultural heritage or manuscript dataset. The distribution is strongly left-skewed (skew = -1.59) with an unusually heavy left tail: while 50% of records fall between 1597 and 1660 (IQR = 63 years), the minimum reaches back to 1300, producing 714 outliers (7.1% of rows) at the early extreme. The high kurtosis (4.32) confirms a sharp central peak around the median of 1630 with fat tails, and an 8.5% null rate warrants attention before temporal analysis. Treatment: Treat as an ordinal/temporal feature; investigate records below ~1500 as potential data-quality issues or deliberate historical outliers before binning into periods for modelling. high · anthropic:default

n

10,940

nulls

931 (8.5%)

unique

430

min

1,300

max

1,850

mean

1621

median

1,630

std

66.25

q1

1,597

q3

1,660

iqr

63

skew

-1.586

kurtosis

4.319

n_outliers

714

outlier_rate

0.07134

zero_rate

0

This column represents a decade or year of origin — likely a composition or publication year — spanning 1300 to 1850, with only 53 distinct values across 10,940 rows. The distribution is heavily left-skewed (skew = −1.48) with high kurtosis (3.85), meaning most records cluster in the 1590–1650 range (IQR = 60 years) while a long tail stretches back to 1300. Notably, 848 rows (≈7.75%) are flagged as outliers, likely corresponding to early-period records far from the central mass near the median of 1620. Treatment: Treat as an ordinal/temporal feature; consider binning into broader periods or applying a robust scaler given the heavy left tail and outlier concentration. high · anthropic:default

n

10,940

nulls

0 (0.0%)

unique

53

min

1,300

max

1,850

mean

1615

median

1,620

std

66.68

q1

1,590

q3

1,650

iqr

60

skew

-1.478

kurtosis

3.85

n_outliers

848

outlier_rate

0.07751

zero_rate

0

This column contains city names, likely representing geographic origin or location associated with records in the dataset. The null rate of 47.65% is a significant concern — nearly half of all 10,940 rows are missing a city value, triggering an alert. With 906 unique cities and an entropy ratio of 0.856, the distribution is fairly broad, yet Geneva dominates with 320 occurrences (5.59% of non-null rows). The top cities — Geneva, Budingen, Bruges, Munich, Augsburg, Venice — suggest a historical European dataset, possibly pre-modern trade, migration, or administrative records. Treatment: Impute or flag nulls (47.65% missing); encode as categorical feature, potentially grouping rare cities below a frequency threshold given 906 unique values. high · anthropic:default

n

10,940

nulls

5,213 (47.7%)

unique

906

top_value

Geneva

top_rate

0.05588

cardinality

906

entropy

8.406

entropy_ratio

0.8557

This column records the country of origin or location for each record, covering 19 distinct countries across 10,940 rows with no nulls. The distribution is heavily concentrated in Western Europe, with the United Kingdom (3,750 rows, 34.3%) and Germany (3,417 rows) together accounting for roughly two-thirds of all records. Switzerland (1,272) and France (807) are the next largest groups, while the remaining 15 countries collectively represent a small tail — Spain, for example, appears only 29 times. The entropy ratio of 0.59 reflects this moderate-to-high imbalance, which could bias any model trained on country as a feature. Treatment: One-hot encode or target-encode; consider grouping low-frequency countries (e.g., Spain with 29 rows) into an 'Other' bucket to reduce sparsity. high · anthropic:default

n

10,940

nulls

0 (0.0%)

unique

19

top_value

United Kingdom

top_rate

0.3428

cardinality

19

entropy

2.502

entropy_ratio

0.5889

This column likely records the number of attempts made for some action (e.g., quiz attempts, login tries, or retry counts), given its name 'tried' and integer-like distribution starting at 1. The distribution is extremely concentrated: Q1, median, and Q3 are all 1.0, yet the mean is ~3.95 and the max is 500, indicating a tiny fraction of records drive nearly all the variance. With 22.5% of rows flagged as outliers and a kurtosis of 316, the tail is extraordinarily heavy and the bulk of users attempt something exactly once. Treatment: Log-transform (log1p) before modelling, or cap at a meaningful percentile threshold to reduce outlier influence; consider binning into ordinal buckets (1, 2–5, 6+). high · anthropic:default

n

10,940

nulls

0 (0.0%)

unique

111

min

1

max

500

mean

3.952

median

1

std

19.26

q1

1

q3

1

iqr

0

skew

15.61

kurtosis

316

n_outliers

2,457

outlier_rate

0.2246

zero_rate

0

This column records death counts per observation (likely per event, location, or time period), with the vast majority of rows — 75.5% — being exactly zero and a median of 0. The distribution is extraordinarily right-skewed (skew=28.52, kurtosis=991.65), driven by rare but extreme values reaching a maximum of 500; notably, 24.5% of rows are flagged as outliers, meaning non-zero death counts are statistically rare but not negligible. Only 74 unique values across 10,940 rows confirms the heavily zero-inflated, discrete nature of the data. Treatment: Model with zero-inflated or negative-binomial regression; apply log1p transform if used as a feature in standard ML pipelines. high · anthropic:default

n

10,940

nulls

0 (0.0%)

unique

74

min

0

max

500

mean

1.493

median

0

std

13.19

q1

0

q3

0

iqr

0

skew

28.52

kurtosis

991.6

n_outliers

2,684

outlier_rate

0.2453

zero_rate

0.7547