pandrs 0.4.1

A high-performance DataFrame library for Rust, providing pandas-like API with advanced features including SIMD optimization, parallel processing, and distributed computing capabilities
Documentation 
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

//! Aggregation helper functions

use crate::core::error::Result;
use crate::dataframe::DataFrame;
use std::collections::HashMap;

/// Compute standard error of the mean for a column
pub fn sem(df: &DataFrame, column: &str, ddof: usize) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let valid_values: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid_values.len() <= ddof {
        return Ok(f64::NAN);
    }

    let n = valid_values.len() as f64;
    let mean = valid_values.iter().sum::<f64>() / n;
    let variance: f64 = valid_values.iter().map(|v| (v - mean).powi(2)).sum::<f64>()
        / (valid_values.len() - ddof) as f64;
    let std_dev = variance.sqrt();

    Ok(std_dev / n.sqrt())
}

/// Compute mean absolute deviation for a column
pub fn mad(df: &DataFrame, column: &str) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let valid_values: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid_values.is_empty() {
        return Ok(f64::NAN);
    }

    let mean = valid_values.iter().sum::<f64>() / valid_values.len() as f64;
    let mad =
        valid_values.iter().map(|v| (v - mean).abs()).sum::<f64>() / valid_values.len() as f64;

    Ok(mad)
}

/// Compute the product of values in a column
pub fn prod(df: &DataFrame, column: &str) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let valid_values: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid_values.is_empty() {
        return Ok(f64::NAN);
    }

    Ok(valid_values.iter().product())
}

/// Get statistics for a single numeric column
///
/// Kept numerically consistent with [`PandasCompatExt::describe`]
/// (`super::super::functions::functions_2_impl_part1::describe`): both use
/// ddof=1 sample variance (undefined -- NaN, not 0 -- for a single
/// observation) and the same linear-interpolation quantile method, so the
/// two don't disagree on the same column.
///
/// [`PandasCompatExt::describe`]: crate::dataframe::pandas_compat::trait_def::PandasCompatExt::describe
pub fn describe_column(df: &DataFrame, column: &str) -> Result<HashMap<String, f64>> {
    let values = df.get_column_numeric_values(column)?;
    let valid: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    let mut result = HashMap::new();

    if valid.is_empty() {
        result.insert("count".to_string(), 0.0);
        result.insert("mean".to_string(), f64::NAN);
        result.insert("std".to_string(), f64::NAN);
        result.insert("min".to_string(), f64::NAN);
        result.insert("max".to_string(), f64::NAN);
        result.insert("25%".to_string(), f64::NAN);
        result.insert("50%".to_string(), f64::NAN);
        result.insert("75%".to_string(), f64::NAN);
        return Ok(result);
    }

    let n = valid.len() as f64;
    let sum: f64 = valid.iter().sum();
    let mean = sum / n;
    // Sample standard deviation (ddof=1, pandas' default): undefined for a
    // single observation. The previous `(n - 1.0).max(1.0)` divisor clamp
    // avoided a division by zero by silently treating n=1 as n=2, which
    // reports a spurious 0 instead of NaN.
    let std = if valid.len() < 2 {
        f64::NAN
    } else {
        let variance: f64 = valid.iter().map(|v| (v - mean).powi(2)).sum::<f64>() / (n - 1.0);
        variance.sqrt()
    };
    let min = valid.iter().cloned().fold(f64::INFINITY, f64::min);
    let max = valid.iter().cloned().fold(f64::NEG_INFINITY, f64::max);

    result.insert("count".to_string(), n);
    result.insert("mean".to_string(), mean);
    result.insert("std".to_string(), std);
    result.insert("min".to_string(), min);
    result.insert("max".to_string(), max);

    // Quartiles via linear interpolation (matches `describe()`'s
    // percentiles, and pandas' default `interpolation="linear"`) instead of
    // naive truncated indexing.
    let mut sorted = valid.clone();
    sorted.sort_by(|a, b| a.total_cmp(b));
    let percentile = |p: f64| -> f64 {
        if sorted.len() == 1 {
            return sorted[0];
        }
        let idx = p * (sorted.len() - 1) as f64;
        let lower = idx.floor() as usize;
        let upper = idx.ceil() as usize;
        if lower == upper {
            sorted[lower]
        } else {
            let weight = idx - lower as f64;
            sorted[lower] * (1.0 - weight) + sorted[upper] * weight
        }
    };

    result.insert("25%".to_string(), percentile(0.25));
    result.insert("50%".to_string(), percentile(0.50));
    result.insert("75%".to_string(), percentile(0.75));

    Ok(result)
}

/// Compute geometric mean for a column
pub fn geometric_mean(df: &DataFrame, column: &str) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let valid_values: Vec<f64> = values
        .iter()
        .filter(|v| !v.is_nan() && **v > 0.0)
        .copied()
        .collect();

    if valid_values.is_empty() {
        return Ok(f64::NAN);
    }

    let log_sum: f64 = valid_values.iter().map(|v| v.ln()).sum();
    Ok((log_sum / valid_values.len() as f64).exp())
}

/// Compute harmonic mean for a column
pub fn harmonic_mean(df: &DataFrame, column: &str) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let valid_values: Vec<f64> = values
        .iter()
        .filter(|v| !v.is_nan() && **v != 0.0)
        .copied()
        .collect();

    if valid_values.is_empty() {
        return Ok(f64::NAN);
    }

    let reciprocal_sum: f64 = valid_values.iter().map(|v| 1.0 / v).sum();
    Ok(valid_values.len() as f64 / reciprocal_sum)
}

/// Compute interquartile range (IQR) for a column
pub fn iqr(df: &DataFrame, column: &str) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let mut valid: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid.len() < 2 {
        return Ok(f64::NAN);
    }

    valid.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));

    let q25_idx = (valid.len() as f64 * 0.25) as usize;
    let q75_idx = (valid.len() as f64 * 0.75) as usize;

    let q25 = valid.get(q25_idx).copied().unwrap_or(f64::NAN);
    let q75 = valid.get(q75_idx).copied().unwrap_or(f64::NAN);

    Ok(q75 - q25)
}

/// Compute coefficient of variation (CV) for a column
pub fn cv(df: &DataFrame, column: &str) -> Result<f64> {
    let values = df.get_column_numeric_values(column)?;
    let valid: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid.is_empty() {
        return Ok(f64::NAN);
    }

    let n = valid.len() as f64;
    let mean = valid.iter().sum::<f64>() / n;

    if mean.abs() < f64::EPSILON {
        return Ok(f64::NAN);
    }

    let variance: f64 = valid.iter().map(|v| (v - mean).powi(2)).sum::<f64>() / (n - 1.0).max(1.0);
    let std = variance.sqrt();

    Ok(std / mean.abs())
}

/// Compute specific percentile for a column
pub fn percentile_value(df: &DataFrame, column: &str, q: f64) -> Result<f64> {
    if q < 0.0 || q > 1.0 {
        return Ok(f64::NAN);
    }

    let values = df.get_column_numeric_values(column)?;
    let mut valid: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid.is_empty() {
        return Ok(f64::NAN);
    }

    valid.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));

    let idx = ((valid.len() - 1) as f64 * q) as usize;
    Ok(valid.get(idx).copied().unwrap_or(f64::NAN))
}

/// Compute trimmed mean (excluding outliers at both ends)
pub fn trimmed_mean(df: &DataFrame, column: &str, trim_fraction: f64) -> Result<f64> {
    if trim_fraction < 0.0 || trim_fraction >= 0.5 {
        return Ok(f64::NAN);
    }

    let values = df.get_column_numeric_values(column)?;
    let mut valid: Vec<f64> = values.iter().filter(|v| !v.is_nan()).copied().collect();

    if valid.is_empty() {
        return Ok(f64::NAN);
    }

    valid.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));

    let trim_count = (valid.len() as f64 * trim_fraction) as usize;
    let trimmed = &valid[trim_count..valid.len() - trim_count];

    if trimmed.is_empty() {
        return Ok(f64::NAN);
    }

    Ok(trimmed.iter().sum::<f64>() / trimmed.len() as f64)
}