quirky bioluminescence

Taxonomic species/genus identifiers (Latin binomials like 'Mnemiopsis leidyi' and genera like 'Lingulodinium', 'Vibrio'). With 245 unique values across 43,060 rows and entropy ratio 0.747, the distribution is moderately spread — the top species accounts for only 4.6% of rows. The mix of binomial species names and bare genus names suggests inconsistent taxonomic resolution across records. Treatment: Treat as a categorical label; consider normalizing genus-only vs. species-level entries before grouping or modelling. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

245

top_value

Mnemiopsis leidyi

top_rate

0.04645

cardinality

245

entropy

5.928

entropy_ratio

0.747

Categorical genus label across 27 bioluminescent marine genera (Noctiluca, Pyrocystis, Alexandrium, Aequorea, etc.). Distribution is essentially uniform — the top 10 genera each show exactly 2000 rows and the top rate is 4.64%, giving an entropy ratio of 0.959, which signals a deliberately balanced sample rather than natural abundance. No nulls across 43,060 rows. Treatment: use directly as the classification target; one-hot or label-encode for modelling. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

27

top_value

Noctiluca

top_rate

0.04645

cardinality

27

entropy

4.558

entropy_ratio

0.9586

Taxonomic family labels for what appears to be a catalogue of bioluminescent marine organisms, spanning 22 distinct families across 43,060 complete rows. The distribution is highly engineered rather than natural: four families (Pyrocystaceae, Euphausiidae, Cypridinidae, Vibrionaceae) each hit exactly 4,000 rows and several others land at exactly 2,000, suggesting deliberate per-class sampling or quota balancing. Entropy ratio of 0.93 confirms the near-uniform spread, and no nulls are present. Treatment: Use directly as a categorical class label; one-hot or integer-encode for modelling. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

22

top_value

Pyrocystaceae

top_rate

0.09289

cardinality

22

entropy

4.157

entropy_ratio

0.9322

Taxonomic phylum label across 43060 records spanning just 7 distinct values with no nulls. Arthropoda leads at 28.6% (12297 rows), followed by Cnidaria (8874) and Myzozoa (8000), with entropy ratio 0.92 indicating a fairly even spread across the seven categories. The mix of animal phyla alongside Proteobacteria (a bacterial phylum) is notable — this column blends kingdoms. Treatment: one-hot encode directly given the low cardinality. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

7

top_value

Arthropoda

top_rate

0.2856

cardinality

7

entropy

2.593

entropy_ratio

0.9235

Taxonomic class labels for marine organisms, spanning 13 distinct values across 43,060 rows with no nulls. Distribution is fairly balanced (entropy ratio 0.93) with Dinophyceae leading at only 18.6% — several classes show suspiciously round counts (8000, 6000, 4000, 2000) suggesting curated/sampled rather than naturally observed frequencies. Treatment: use directly as a multi-class classification target with label encoding. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

13

top_value

Dinophyceae

top_rate

0.1858

cardinality

13

entropy

3.43

entropy_ratio

0.9268

Taxonomic order names for marine organisms (Gonyaulacales, Euphausiacea, Calanoida, etc.), with 17 distinct values across 43,060 rows and no nulls. Distribution is fairly even — entropy ratio 0.949 and top class only 13.9% — though several categories sit at suspiciously round counts (4000, 2000), suggesting stratified sampling or quota construction rather than natural frequencies. Treatment: one-hot or target-encode; safe to use directly given low cardinality and no missing values. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

17

top_value

Gonyaulacales

top_rate

0.1393

cardinality

17

entropy

3.879

entropy_ratio

0.9491

Numeric column bounded between -76.619 and 88.29, consistent with WGS84 latitudes in degrees. The distribution is wide (std 40.27, IQR 69.61) and mildly left-skewed (-0.66) with a flat shape (kurtosis -0.94), indicating coverage across both hemispheres rather than a single region. No nulls and no outliers flagged across 43,060 rows with 14,146 distinct values. Treatment: Pair with longitude for geospatial features; consider binning or projecting rather than using raw degrees in a linear model. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

14,146

min

-76.62

max

88.29

mean

19.1

median

36.71

std

40.27

q1

-19.31

q3

50.3

iqr

69.61

skew

-0.6614

kurtosis

-0.9355

n_outliers

0

outlier_rate

0

zero_rate

0.0004645

Geographic longitude in decimal degrees, spanning the full -179.9987 to 179.99 range with 14,637 distinct values across 43,060 rows and zero nulls. The distribution is nearly symmetric (skew 0.14) with light tails (kurtosis -0.65) and a wide IQR of 124.12, indicating truly global coverage rather than a regional sample. No outliers flagged and only a 0.11% zero rate, consistent with clean coordinate data. Treatment: Pair with latitude for geospatial features; avoid treating as a linear scalar since ±180 wraps. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

14,637

min

-180

max

180

mean

9.64

median

3.057

std

88.61

q1

-60.19

q3

63.93

iqr

124.1

skew

0.1376

kurtosis

-0.6464

n_outliers

0

outlier_rate

0

zero_rate

0.001115

This is a numeric depth measurement (likely meters), with 24.75% nulls across 43,060 rows and only 3,283 distinct values. The distribution is heavily right-skewed (skew 4.72, kurtosis 35.89): the median is 52.5 but the mean is 281.2 and the max reaches 10,000, while 10.63% of values are flagged as outliers. Notably, the minimum is -53.0 (negative depths are suspect) and 11.92% of values are exactly zero. Treatment: Investigate negative and zero values, then log-transform (after shifting) before modelling. high · anthropic:claude-opus-4-7

n

43,060

nulls

10,658 (24.8%)

unique

3,283

min

-53

max

10,000

mean

281.2

median

52.5

std

570.2

q1

7.5

q3

321

iqr

313.5

skew

4.724

kurtosis

35.89

n_outliers

3,444

outlier_rate

0.1063

zero_rate

0.1192

This is a date column stored as free text rather than a parsed timestamp, with values mixing single dates (e.g. '2017-05-30'), single months ('2013-08'), month ranges ('2010-05/2010-06') and even multi-year spans ('1962/1964'). The format heterogeneity is the main surprise: 97% of entries are one 'word', but length varies from 4 to 51 characters, and 67% are duplicates of another row. Nulls are also non-trivial at 12%. Treatment: Parse into structured start/end dates (handling year-only, month, and range formats) before any temporal analysis. high · anthropic:claude-opus-4-7

n

43,060

nulls

5,182 (12.0%)

unique

12,338

len_min

4

len_max

51

len_mean

16.45

len_median

19

len_p95

39

word_mean

1.03

word_median

1

n_empty

0

n_duplicates

25,540

duplicate_rate

0.6743

vocab_size

10,135

readability_flesch_mean

121.2

emoji_rate

0

url_rate

0

one_word_rate

0.9705

allcaps_rate

1

boilerplate_rate

0

This is a year column stored categorically across 137 distinct values, suggesting coverage spanning over a century. The most common year is '2000' at 5.17% of non-null rows, with a high entropy ratio of 0.865 indicating values are spread fairly evenly across years. Notably, 42.18% of rows are null, which triggered a null_rate alert and limits usefulness without imputation or filtering. Treatment: Cast to integer year and decide whether to drop or impute the 42% missing rows before time-based analysis. high · anthropic:claude-opus-4-7

n

43,060

nulls

18,164 (42.2%)

unique

137

top_value

2000

top_rate

0.0517

cardinality

137

entropy

6.142

entropy_ratio

0.8653

Country of origin as a categorical label across 130 distinct values, but 63.7% of the 43,060 rows are empty strings rather than nulls, making the modal 'value' effectively missing. The remaining entries show inconsistent encoding — full names ('Australia', 'United States'), ISO codes ('GB'), uppercase forms ('PERU', 'SOVIET UNION'), and a defunct state — suggesting data was merged from heterogeneous sources without normalisation. Entropy ratio of 0.37 confirms the distribution is heavily concentrated in a few buckets. Treatment: Normalise to ISO country codes, treat empty string as missing, then one-hot or target-encode. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

130

top_value

top_rate

0.6368

cardinality

130

entropy

2.569

entropy_ratio

0.3658

This column names the source dataset each record was drawn from, with 214 distinct provenance strings across 43,060 rows. The dominant value is an empty string covering 61.1% of rows (26,317), meaning provenance is missing for the majority; named sources like 'Environmental Monitoring database (MOD) DNV' (1,760) and 'Jellyfish sightings along the Italian coastline from 2009 to 2017' (1,024) trail far behind. Entropy ratio of 0.41 confirms the distribution is heavily concentrated on that blank. Treatment: Treat empty string as missing and group rare sources before any per-dataset stratification. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

214

top_value

top_rate

0.6112

cardinality

214

entropy

3.19

entropy_ratio

0.4121

Categorical taxonomy label grouping records by bioluminescent organism type, with 26 distinct groups across 43,060 rows and no nulls. Distribution is remarkably flat (entropy ratio 0.95): 'Dinoflagellate' leads at only 9.3%, and the next nine values are tied at exactly 2,000 rows each, suggesting a synthetic or quota-balanced sample rather than naturally observed frequencies. Treatment: One-hot or target-encode; the suspiciously uniform 2,000-per-class counts warrant a check for synthetic balancing before modelling. high · anthropic:claude-opus-4-7

n

43,060

nulls

0 (0.0%)

unique

26

top_value

Dinoflagellate

top_rate

0.09289

cardinality

26

entropy

4.465

entropy_ratio

0.95