1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
//! Data-quality checks derived from a [`DatasetProfile`].
//!
//! Each [`QualityIssue`] is a single human-readable finding with a severity.
//! The thresholds are conservative defaults; callers may filter the returned
//! list as desired.
use crate::profile::DatasetProfile;
use crate::types::{ColumnType, Severity};
/// The category of a [`QualityIssue`].
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
#[cfg_attr(feature = "serde", derive(serde::Serialize))]
pub enum QualityKind {
/// A column has a missing fraction at or above the threshold.
HighMissing,
/// A numeric column has (near-)zero variance — it carries no signal.
ConstantColumn,
/// A categorical column has cardinality equal to the row count (likely an
/// identifier rather than a feature).
NearUnique,
/// The dataset contains exact-duplicate rows.
DuplicateRows,
/// A numeric column has values outside the Tukey IQR fences.
Outliers,
/// A categorical column is dominated by a single value.
Imbalance,
/// A pair of numeric columns is highly correlated (`|r| >= threshold`).
HighCorrelation,
/// A feature column is suspiciously highly correlated with the target column.
TargetLeakage,
}
/// A single data-quality finding.
#[derive(Debug, Clone, PartialEq)]
#[cfg_attr(feature = "serde", derive(serde::Serialize))]
pub struct QualityIssue {
/// What kind of issue this is.
pub kind: QualityKind,
/// How serious the issue is.
pub severity: Severity,
/// Which column the issue concerns, or `None` for dataset-wide findings.
pub column: Option<String>,
/// Human-readable description, suitable for direct display in a report.
pub message: String,
}
/// Thresholds controlling when each check fires.
///
/// All fields are intentionally `pub` so callers can tune them before running
/// [`run_checks`].
#[derive(Debug, Clone, Copy, PartialEq)]
#[cfg_attr(feature = "serde", derive(serde::Serialize))]
pub struct Thresholds {
/// Missing fraction at or above which [`QualityKind::HighMissing`] fires.
pub missing_fraction: f64,
/// Variance at or below which [`QualityKind::ConstantColumn`] fires.
pub near_zero_variance: f64,
/// `unique / n_rows` at or above which [`QualityKind::NearUnique`] fires.
pub near_unique_ratio: f64,
/// Outlier fraction at or above which [`QualityKind::Outliers`] fires.
pub outlier_fraction: f64,
/// Imbalance ratio (`freq / present`) at or above which
/// [`QualityKind::Imbalance`] fires.
pub imbalance_ratio: f64,
/// Correlation magnitude `|r|` at or above which [`QualityKind::HighCorrelation`] fires.
pub high_correlation: f64,
/// Correlation magnitude `|r|` or Cramér's V at or above which [`QualityKind::TargetLeakage`] fires.
pub target_leakage: f64,
}
impl Default for Thresholds {
fn default() -> Self {
Thresholds {
missing_fraction: 0.5,
near_zero_variance: 1e-12,
near_unique_ratio: 0.98,
outlier_fraction: 0.05,
imbalance_ratio: 0.95,
high_correlation: 0.95,
target_leakage: 0.90,
}
}
}
/// Runs all data-quality checks against `profile` using `thresholds`.
pub fn run_checks(profile: &DatasetProfile, thresholds: &Thresholds) -> Vec<QualityIssue> {
let mut issues = Vec::new();
for col in &profile.columns {
if col.missing_fraction >= thresholds.missing_fraction && col.count > 0 {
issues.push(QualityIssue {
kind: QualityKind::HighMissing,
severity: if col.missing_fraction >= 0.9 {
Severity::Critical
} else {
Severity::Warning
},
column: Some(col.name.clone()),
message: format!(
"{}: {:.1}% of values are missing",
col.name,
col.missing_fraction * 100.0
),
});
}
match col.column_type {
ColumnType::Numeric => {
if let Some(n) = &col.numeric {
let var = n.std * n.std;
if var <= thresholds.near_zero_variance {
issues.push(QualityIssue {
kind: QualityKind::ConstantColumn,
severity: Severity::Warning,
column: Some(col.name.clone()),
message: format!(
"{}: near-zero variance ({:.3e}); column is effectively constant",
col.name, var
),
});
}
if n.outlier_count > 0 && n.outlier_fraction >= thresholds.outlier_fraction {
issues.push(QualityIssue {
kind: QualityKind::Outliers,
severity: if n.outlier_fraction >= 0.2 {
Severity::Warning
} else {
Severity::Info
},
column: Some(col.name.clone()),
message: format!(
"{}: {} outliers ({:.1}%) beyond IQR fences",
col.name,
n.outlier_count,
n.outlier_fraction * 100.0
),
});
}
}
}
ColumnType::Categorical => {
if let Some(c) = &col.categorical {
if col.count > 0 {
let ratio = c.unique as f64 / col.count as f64;
if ratio >= thresholds.near_unique_ratio {
issues.push(QualityIssue {
kind: QualityKind::NearUnique,
severity: Severity::Info,
column: Some(col.name.clone()),
message: format!(
"{}: {} unique values across {} rows (ratio {:.2}); likely an identifier",
col.name, c.unique, col.count, ratio
),
});
}
if c.imbalance_ratio >= thresholds.imbalance_ratio {
issues.push(QualityIssue {
kind: QualityKind::Imbalance,
severity: Severity::Critical,
column: Some(col.name.clone()),
message: format!(
"{}: top value '{}' covers {:.1}% of rows",
col.name,
c.top,
c.imbalance_ratio * 100.0
),
});
}
}
}
}
}
}
if profile.duplicate_rows > 0 {
issues.push(QualityIssue {
kind: QualityKind::DuplicateRows,
severity: if profile.duplicate_fraction >= 0.1 {
Severity::Warning
} else {
Severity::Info
},
column: None,
message: format!(
"{} of {} rows are exact duplicates ({:.2}%)",
profile.duplicate_rows,
profile.n_rows,
profile.duplicate_fraction * 100.0
),
});
}
// Check relationship findings (HighCorrelation & TargetLeakage)
if let Some(rels) = &profile.relationships {
// Pearson high correlation
if let Some(pearson) = &rels.pearson {
let p = pearson.labels.len();
for i in 0..p {
for j in (i + 1)..p {
let r = pearson.values[i][j];
let abs_r = r.abs();
// Check high correlation between feature pairs
if abs_r >= thresholds.high_correlation {
issues.push(QualityIssue {
kind: QualityKind::HighCorrelation,
severity: Severity::Warning,
column: Some(pearson.labels[i].clone()),
message: format!(
"High Pearson correlation between '{}' and '{}' (r = {:.3})",
pearson.labels[i], pearson.labels[j], r
),
});
}
// Check target leakage if target_column matches either column
if let Some(target) = &profile.target_column {
let is_i_target = &pearson.labels[i] == target;
let is_j_target = &pearson.labels[j] == target;
if (is_i_target || is_j_target)
&& !(is_i_target && is_j_target)
&& abs_r >= thresholds.target_leakage
{
let feature = if is_i_target {
&pearson.labels[j]
} else {
&pearson.labels[i]
};
issues.push(QualityIssue {
kind: QualityKind::TargetLeakage,
severity: Severity::Critical,
column: Some(feature.clone()),
message: format!(
"Suspected target leakage: feature '{}' has strong correlation with target '{}' (r = {:.3})",
feature, target, r
),
});
}
}
}
}
}
// Cramér's V target leakage & high correlation
if let Some(cramers) = &rels.cramers_v {
let p = cramers.labels.len();
for i in 0..p {
for j in (i + 1)..p {
let v = cramers.values[i][j];
if let Some(target) = &profile.target_column {
let is_i_target = &cramers.labels[i] == target;
let is_j_target = &cramers.labels[j] == target;
if (is_i_target || is_j_target)
&& !(is_i_target && is_j_target)
&& v >= thresholds.target_leakage
{
let feature = if is_i_target {
&cramers.labels[j]
} else {
&cramers.labels[i]
};
issues.push(QualityIssue {
kind: QualityKind::TargetLeakage,
severity: Severity::Critical,
column: Some(feature.clone()),
message: format!(
"Suspected target leakage: categorical feature '{}' has high Cramér's V with target '{}' (V = {:.3})",
feature, target, v
),
});
}
}
}
}
}
// Point-biserial target leakage
if let Some(target) = &profile.target_column {
for pb in &rels.point_biserial {
let abs_r = pb.correlation.abs();
if abs_r >= thresholds.target_leakage {
let is_cat_target = &pb.categorical == target;
let is_num_target = &pb.numeric == target;
if (is_cat_target || is_num_target) && !(is_cat_target && is_num_target) {
let feature = if is_cat_target {
&pb.numeric
} else {
&pb.categorical
};
issues.push(QualityIssue {
kind: QualityKind::TargetLeakage,
severity: Severity::Critical,
column: Some(feature.clone()),
message: format!(
"Suspected target leakage: feature '{}' has high point-biserial correlation with target '{}' (r = {:.3})",
feature, target, pb.correlation
),
});
}
}
}
}
}
issues
}