matten_data/
numeric.rs

1//! Explicit numeric conversion to a [`matten::Tensor`] (RFC-035 §7–8).
2//!
3//! Conversion is always explicit (`try_numeric()` then `to_tensor()`); there is no
4//! silent coercion. Missing values never become zero, booleans never become 1/0,
5//! and non-numeric text is rejected.
6
7use crate::error::MattenDataError;
8use crate::table::{CellValue, Table};
9
10/// A table whose cells have all been validated as numeric (RFC-034 §4.4).
11///
12/// Produced by [`Table::try_numeric`]. Holds row-major `f64` data ready to become
13/// a [`matten::Tensor`] via [`NumericTable::to_tensor`].
14#[derive(Debug, Clone)]
15pub struct NumericTable {
16    headers: Vec<String>,
17    data: Vec<f64>,
18    rows: usize,
19    cols: usize,
20}
21
22impl NumericTable {
23    /// Number of rows.
24    pub fn row_count(&self) -> usize {
25        self.rows
26    }
27
28    /// Number of columns.
29    pub fn column_count(&self) -> usize {
30        self.cols
31    }
32
33    /// Column names, in column order.
34    pub fn column_names(&self) -> &[String] {
35        &self.headers
36    }
37
38    /// Build a numeric [`matten::Tensor`] with shape `[rows, columns]`, row-major.
39    ///
40    /// Errors with [`MattenDataError::EmptySelection`] if there are no columns, and
41    /// wraps any core construction failure (for example a zero-length dimension
42    /// when there are no rows) as [`MattenDataError::Matten`].
43    pub fn to_tensor(&self) -> Result<matten::Tensor, MattenDataError> {
44        if self.cols == 0 {
45            return Err(MattenDataError::EmptySelection);
46        }
47        matten::Tensor::try_new(self.data.clone(), &[self.rows, self.cols])
48            .map_err(MattenDataError::Matten)
49    }
50}
51
52/// Strictly convert a table's cells to `f64` (RFC-035 §7.1–7.4).
53pub(crate) fn try_numeric(table: &Table) -> Result<NumericTable, MattenDataError> {
54    let headers: Vec<String> = table.headers().to_vec();
55    let cols = headers.len();
56    let table_rows = table.rows();
57    let rows = table_rows.len();
58
59    let mut data = Vec::with_capacity(rows * cols);
60    for (r, row) in table_rows.iter().enumerate() {
61        // One-based CSV line number: header is line 1, first data row is line 2.
62        let line = r + 2;
63        for (c, cell) in row.iter().enumerate() {
64            let value = match cell {
65                CellValue::Int(i) => *i as f64,
66                CellValue::Float(f) => *f,
67                CellValue::Missing => {
68                    return Err(MattenDataError::MissingValue {
69                        column: headers[c].clone(),
70                        row: line,
71                    });
72                }
73                CellValue::Bool(b) => {
74                    return Err(MattenDataError::NonNumericValue {
75                        column: headers[c].clone(),
76                        row: line,
77                        value: b.to_string(),
78                    });
79                }
80                CellValue::Text(s) => {
81                    return Err(MattenDataError::NonNumericValue {
82                        column: headers[c].clone(),
83                        row: line,
84                        value: s.clone(),
85                    });
86                }
87            };
88            data.push(value);
89        }
90    }
91
92    Ok(NumericTable {
93        headers,
94        data,
95        rows,
96        cols,
97    })
98}
matten_data/numeric.rs

matten_data/
numeric.rs
