quirky shipwrecks

This column holds short text labels for individual records, almost certainly vessel or wreck names: 5,497 of 5,569 values are unique, mean length is 17.8 characters with a median of 2 words, and the dominant token "shipwreck" appears 3,706 times alongside nautical prefixes like "ss", "uss", and "hms". Despite the near_unique alert, there are 72 duplicates (1.3%) worth inspecting, and the recurring "shipwreck"/"(wrack)" tokens suggest names follow a templated pattern rather than being free prose. Treatment: Treat as a name identifier; strip boilerplate tokens like "shipwreck" before any text matching, and do not use as a model feature. high · anthropic:claude-opus-4-7

n

5,569

nulls

0 (0.0%)

unique

5,497

len_min

2

len_max

153

len_mean

17.81

len_median

20

len_p95

21

word_mean

2.073

word_median

2

n_empty

0

n_duplicates

72

duplicate_rate

0.01293

vocab_size

6,255

readability_flesch_mean

71.33

emoji_rate

0

url_rate

0

one_word_rate

0.1034

allcaps_rate

0.01724

boilerplate_rate

0

This is a latitude feature spanning -77.42 to 82.17, covering nearly the full geographic range from Antarctica to the high Arctic. The distribution is left-skewed (skew -1.14) with a median of 40.64 well above the mean of 28.41, suggesting a concentration of points in northern mid-latitudes with a tail of southern hemisphere observations. With 5561 unique values across 5569 rows and no nulls, each record carries a near-distinct coordinate. Treatment: Pair with the matching longitude column for geospatial features rather than using as a standalone scalar. high · anthropic:claude-opus-4-7

n

5,569

nulls

0 (0.0%)

unique

5,561

min

-77.42

max

82.17

mean

28.41

median

40.64

std

31.36

q1

12.54

q3

50.36

iqr

37.82

skew

-1.136

kurtosis

0.08299

n_outliers

112

outlier_rate

0.02011

zero_rate

0

This column holds longitude coordinates, with values ranging from -179.28 to 179.45 spanning the full globe and 5568 unique values across 5569 rows. The distribution is mildly right-skewed (0.53) with a median of 2.03 sitting near the prime meridian, and the IQR of 80.73 suggests broad geographic coverage. The flagged 542 outliers (9.7%) likely reflect points in the Pacific tails rather than data errors, given valid lon bounds. Treatment: Pair with latitude as a geospatial feature; avoid treating outliers as anomalies since extremes are valid longitudes. high · anthropic:claude-opus-4-7

n

5,569

nulls

0 (0.0%)

unique

5,568

min

-179.3

max

179.4

mean

1.413

median

2.033

std

76.75

q1

-58.6

q3

22.13

iqr

80.73

skew

0.5273

kurtosis

0.2435

n_outliers

542

outlier_rate

0.09732

zero_rate

0

This column records the year (or fuller date) a vessel sank, but it's almost entirely empty — 99.34% null with only 36 distinct values across 5569 rows. Date formats are inconsistent: bare years like '1942' and '1435', ISO strings like '1937-09-02', verbose forms like 'June 7, 1928', and even ranges like '1643..1663' coexist. Entropy ratio of 0.997 confirms the few populated values are nearly all unique, with '1942' the only repeat (2 occurrences). Treatment: Parse to a normalized year integer and treat as sparse metadata; too null-heavy to use as a feature. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,532 (99.3%)

unique

36

top_value

1942

top_rate

0.05405

cardinality

36

entropy

5.155

entropy_ratio

0.9972

Categorical type label for each record, dominated overwhelmingly by maritime wreckage: 'shipwreck' accounts for 5081 of 5569 rows (91.2% top_rate) with 30 distinct values total. Entropy ratio of 0.115 confirms the long_tail alert — the remaining 29 categories split fewer than 500 rows, with several near-synonyms ('ship'/'boat'/'schooner', 'aircraft'/'plane', 'vehicle'/'motor_vehicle') suggesting inconsistent labelling that could be consolidated. Treatment: Collapse synonymous categories and consider binarising as shipwreck-vs-other given the extreme imbalance. high · anthropic:claude-opus-4-7

n

5,569

nulls

0 (0.0%)

unique

30

top_value

shipwreck

top_rate

0.9124

cardinality

30

entropy

0.565

entropy_ratio

0.1151

This column holds Wikipedia article references prefixed with a language code (e.g., 'en:SS Edmund Fitzgerald', 'fr:Armorique (navire)', 'ar:...'), likely linking each record to an encyclopedia entry about a ship, aircraft, or wreck. It is overwhelmingly sparse — 94.38% null with only 307 distinct values across 5569 rows — and the distribution is nearly flat (entropy ratio 0.998, top value appears just 4 times, top_rate 1.28%). The presence of multiple language prefixes (en, fr, ar) signals a mixed-language reference field rather than a clean categorical. Treatment: Treat as an optional external reference link; drop for modelling or split off the language prefix if needed. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,256 (94.4%)

unique

307

top_value

en:SS Edmund Fitzgerald

top_rate

0.01278

cardinality

307

entropy

8.245

entropy_ratio

0.998

This column holds Wikidata Q-identifiers (e.g., Q1286267), linking rows to entities in the Wikidata knowledge graph. It is overwhelmingly sparse — 93.52% null — and among the 5569 rows only 353 unique values appear, with the most common identifier showing up just 4 times (top_rate 0.011). Entropy ratio of 0.998 confirms the non-null values are nearly all distinct, consistent with a foreign key rather than a categorical feature. Treatment: Left-join on this id to enrich with Wikidata attributes; do not use as a model feature directly. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,208 (93.5%)

unique

353

top_value

Q1286267

top_rate

0.01108

cardinality

353

entropy

8.446

entropy_ratio

0.9979

Free-text descriptive notes about wrecks, barges, and other maritime features, populated for only ~6% of rows (null_rate 0.9386). Among the 342 non-null entries there are 291 distinct strings with entropy_ratio 0.975, so values are nearly all unique short narratives; the modal phrase 'WWII era concrete fuel barge converted into breakwater' appears just 14 times (top_rate 0.041). Mixed languages are present (e.g., French 'Chaloupe abandonnée à terre' alongside English), confirming this is curator-authored prose rather than a controlled vocabulary. Treatment: Treat as sparse free text; tokenize/embed for search or keyword extraction rather than using as a categorical feature. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,227 (93.9%)

unique

291

top_value

WWII era concrete fuel barge converted into breakwater

top_rate

0.04094

cardinality

291

entropy

7.983

entropy_ratio

0.9753

A categorical 'heritage' field that is effectively empty: 99.77% null, with only 13 non-null values across 4 distinct levels. The observed values are inconsistent ('2', '1', 'yes', 'no'), suggesting a coding scheme that was never standardized or fully populated. Treatment: Drop; null_rate of 0.9977 leaves too little signal and the value coding is inconsistent. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,556 (99.8%)

unique

4

top_value

2

top_rate

0.7692

cardinality

4

entropy

1.145

entropy_ratio

0.5726

This is an OpenStreetMap-style 'access' tag indicating who may use a feature, with values like 'yes', 'no', 'permit', 'private', 'permissive', 'customers', and 'foot'. It is overwhelmingly null (90.88%), and among the 508 populated rows 'yes' dominates at 66.93%, leaving the other 7 categories thinly represented. Cardinality is only 8 with entropy ratio 0.55, so signal beyond presence/absence is limited. Treatment: Collapse rare levels and encode as a low-cardinality categorical, or reduce to a populated/'yes' indicator given the 90.88% null rate. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,061 (90.9%)

unique

8

top_value

yes

top_rate

0.6693

cardinality

8

entropy

1.649

entropy_ratio

0.5497

A free-text 'depth' field, almost certainly a measurement (likely meters) but stored as strings with mixed formats — bare numbers like '7', '16', '14' coexist with unit-suffixed values like '30m', '25m', and decimals like '12.2'. It is overwhelmingly missing (null_rate 0.9627) and extremely diffuse among the 208 populated rows: 154 unique values, top value '7' covers only 2.88%, and entropy_ratio 0.975 indicates a near-uniform long tail. Treatment: Strip unit suffixes and parse to numeric meters; given >96% nulls, treat as low-signal and consider dropping or flagging presence only. high · anthropic:claude-opus-4-7

n

5,569

nulls

5,361 (96.3%)

unique

154

top_value

7

top_rate

0.02885

cardinality

154

entropy

7.085

entropy_ratio

0.975

Categorical seamark classification with 10 distinct values, almost entirely dominated by 'wreck' at 98.36% of non-null rows (5026 of 5569). The remaining categories are extreme long-tail (hulk at 56, then single-digit counts down to one), and 8.24% of rows are null. Entropy ratio of 0.044 confirms the column carries almost no discriminative signal. Treatment: Drop or collapse to a binary 'is_wreck' flag; near-constant. high · anthropic:claude-opus-4-7

n

5,569

nulls

459 (8.2%)

unique

10

top_value

wreck

top_rate

0.9836

cardinality

10

entropy

0.1477

entropy_ratio

0.04447

This is almost certainly the OpenStreetMap object id: every one of the 5569 rows is unique, no nulls, no zeros, and values span 13M to 13.5B which matches OSM's monotonically growing id space. The distribution is right-skewed (skew 1.07) with mean 4.03B well above the median 2.35B, reflecting OSM's accumulation of newer, higher ids over time rather than anything analytically meaningful. Treatment: Drop from modelling; retain as a join key to OSM source data. high · anthropic:claude-opus-4-7

n

5,569

nulls

0 (0.0%)

unique

5,569

min

1.306e+07

max

1.354e+10

mean

4.032e+09

median

2.349e+09

std

3.875e+09

q1

1.181e+09

q3

6.516e+09

iqr

5.335e+09

skew

1.071

kurtosis

-0.2044

n_outliers

0

outlier_rate

0

zero_rate

0

This column records the OpenStreetMap geometry type for each row, taking only two values: "node" (3656 rows, 65.6%) and "way" (1913 rows). With cardinality 2 and entropy ratio 0.928, the split is fairly balanced but tilted toward nodes, and there are no nulls across all 5569 rows. Treatment: Encode as a binary indicator (node vs way) for modelling. high · anthropic:claude-opus-4-7

n

5,569

nulls

0 (0.0%)

unique

2

top_value

node

top_rate

0.6565

cardinality

2

entropy

0.9281

entropy_ratio

0.9281