Troubleshooting

Symptom → cause → fix, for the messages you actually hit. Every message below is copied verbatim from the source (verified by grep against crates/matten/src) — if what you’re seeing doesn’t match one of these exactly, it isn’t the same problem.

“no associated function or constant named from_csv found for struct Tensor”

error[E0599]: no associated function or constant named `from_csv` found for struct `Tensor`

The csv (or json, for from_json) feature is off. This happens on the lean profile:

matten = { version = "0.48.0", default-features = false }

Either drop default-features = false to get the default profile (serde + json + csv), or opt back in explicitly:

matten = { version = "0.48.0", features = ["csv"] }

See Quick start and Cargo features.

“matten shape error in …”

matten shape error in try_new: data length 3 does not match shape [2, 2], which requires 4 elements

The data you passed doesn’t have exactly shape.iter().product() elements. Recompute the shape from the data’s actual length, or recompute the data from the shape you intend — the message names both numbers so you can tell which one is wrong. See Data model and lifecycle.

“matten broadcast error in …”

matten broadcast error in add: shapes [2, 3] and [2] are not compatible

The two shapes can’t broadcast against each other. Shapes are compared right-aligned: dimensions must be equal, or one of them must be 1. A trailing [2] only broadcasts against another axis of size 2 or 1 at the same right-aligned position — [2, 3] and [2] line up 3 against 2, which is neither equal nor 1. See Operators and broadcasting for the full alignment rule, and use try_add/try_sub/try_mul/try_div if you’d rather get a Result back than have this panic.

“matten unsupported error in …”

matten unsupported error in clip: clip is not supported on dynamic tensors; call try_numeric() first

A numeric-only API (arithmetic, elementwise math, matmul, reductions, …) was called on a dynamic tensor (the dynamic feature’s heterogeneous Element engine). Convert first:

let numeric = dynamic_tensor.try_numeric()?;

See Dynamic feature (Element model).

“matten allocation error: …”

matten allocation error: try_zeros requested 4000000 elements, exceeding the limit of 1048576
(MattenLimits::max_elements); use smaller shapes (requested 4000000 elements)

The requested shape exceeds MattenLimits::max_elements (default 1,048,576 — about 8 MiB of f64). This is a safety limit against runaway allocations from caller-supplied shapes; it does not apply to ordinary operations on tensors already in memory (arithmetic, reductions, slicing). If your shape is legitimately larger, pass a custom MattenLimits to the _with_limits form of the constructor you’re calling (try_zeros_with_limits, try_ones_with_limits, try_full_with_limits).