Interface SparseMatrixInput

Chooses how the one-dimensional arrays encode the two-dimensional matrix.

• "csc" (compressed sparse column): coefficients are grouped by column. starts[j]..starts[j + 1] locates column j, and each corresponding indices[k] is a row index.
• "csr" (compressed sparse row): coefficients are grouped by row. starts[i]..starts[i + 1] locates row i, and each corresponding indices[k] is a column index.

This changes the meanings and required length of both starts and indices; it does not transpose the mathematical matrix. Use "csc" when constructing columns naturally and "csr" when constructing rows naturally.

Minor-axis coordinate for every packed coefficient in values. indices[k] and values[k] always describe the same matrix entry.

In CSC, packed position k belongs to the column determined by starts, and indices[k] is its zero-based row. In CSR, position k belongs to the row determined by starts, and indices[k] is its zero-based column. Consequently CSC indices must be below numRows, while CSR indices must be below numCols.

indices.length must equal values.length and the final starts entry. A single packed row/column must not repeat a minor index, because that would specify the same matrix coordinate twice. Indices in different packed rows/columns may naturally repeat.

For CSC starts: [0, 2, 3], indices: [0, 1, 1] means column 0 has entries in rows 0 and 1, while column 1 has one entry in row 1.

Number of columns in the mathematical matrix. It must be a non-negative signed 32-bit integer.

In CSR format, every entry of indices is a column index and must satisfy 0 <= indices[k] < numCols. In CSC format, this is the number of packed column segments, so starts.length must equal numCols + 1.

Number of rows in the mathematical matrix. It must be a non-negative signed 32-bit integer.

In CSC format, every entry of indices is a row index and must satisfy 0 <= indices[k] < numRows. In CSR format, this is the number of packed row segments, so starts.length must equal numRows + 1.

Boundaries of the packed rows or columns inside indices and values. These are offsets, not model row/column indices.

In CSC, column j occupies positions k from starts[j] inclusive to starts[j + 1] exclusive. In CSR, the same formula describes row j. Therefore an empty row/column has equal adjacent offsets.

The array must have numCols + 1 entries for CSC or numRows + 1 entries for CSR. It must start with 0, be nondecreasing, and end with the exact number of stored entries: starts[starts.length - 1] === indices.length === values.length.

Example: starts: [0, 2, 2, 3] describes three major-axis items. The first has two entries at packed positions 0 and 1, the second is empty, and the third has one entry at position 2.

Numerical matrix coefficient at every packed position. values[k] is at the minor coordinate indices[k] within the row/column selected by starts; all three arrays must therefore be interpreted together.

values.length must equal indices.length and the final starts entry. Values must be numbers other than NaN; matrix coefficients should be finite rather than using infinity to represent a bound. Explicit zero and coefficients whose magnitude is at or below small_matrix_value may be removed by HiGHS, so the later extracted matrix can contain fewer entries.

For the CSC encoding starts: [0, 2, 3], indices: [0, 1, 1], and values: [1, 2, 3], the stored coordinates are (row 0, col 0) = 1, (row 1, col 0) = 2, and (row 1, col 1) = 3.