doc2sequence — Convert tokenized documents to word-vector or word-index sequences.

doc2sequence(emb, documents) returns a cell array whose i-th cell contains the embedding-vector sequence for the i-th tokenized document. doc2sequence(enc, documents) returns word-index sequences from a wordEncoding object.

Syntax

sequences = doc2sequence(emb, documents)
sequences = doc2sequence(enc, documents)
sequences = doc2sequence(emb, documents, Name, Value)
sequences = doc2sequence(enc, documents, Name, Value)

Inputs

Returns

Errors

How doc2sequence works

• The first input must be a RunMat wordEmbedding or wordEncoding compatibility object.
• documents must be a RunMat tokenizedDocument compatibility object.
• For embedding input, each output cell contains a D-by-S numeric matrix, where D is the embedding dimension and S is the sequence length.
• For encoding input, each output cell contains a 1-by-S numeric vector of 1-based word encoding indices.
• Unknown words are discarded by default. Set UnknownWord to "nan" to include NaN vector columns for embedding inputs or NaN index elements for encoding inputs.
• PaddingDirection accepts "left", "right", and "none". The default is left padding.
• PaddingValue accepts a numeric scalar and defaults to 0. MATLAB warns against NaN padding for deep learning workflows, but RunMat accepts numeric NaN for compatibility.
• Length accepts "longest", "shortest", or a positive integer. Positive integer lengths truncate sequences on the right.
• When PaddingDirection is "none", output cells keep per-document sequence lengths, except that fixed or shortest Length values still truncate longer documents.
• Very large padded dense outputs are rejected with a controlled error to avoid accidental memory explosions.

GPU memory and residency

doc2sequence performs host text/object sequence construction and has no provider kernel.

Examples

Convert Documents To Word Vectors

emb = fastTextWordEmbedding;
documents = tokenizedDocument(["alpha beta"; "beta"]);
sequences = doc2sequence(emb, documents)

Expected output:

`sequences` is a cell array of D-by-S matrices.

Convert Documents To Word Indices

documents = tokenizedDocument(["alpha beta"; "beta"]);
enc = wordEncoding(documents);
sequences = doc2sequence(enc, documents)

Expected output:

`sequences` is a cell array of 1-by-S index vectors.

Avoid Padding

sequences = doc2sequence(emb, documents, "PaddingDirection", "none")

Expected output:

Each cell keeps the sequence length of the corresponding document.

Control Padding And Unknown Words

sequences = doc2sequence(emb, documents, "UnknownWord", "nan", "PaddingDirection", "right", "PaddingValue", -1, "Length", 100)

Expected output:

Sequences are right-padded or right-truncated to length 100.

Using doc2sequence with coding agents

Open a RunMat example with live inputs, then ask the agent to explain how doc2sequence changes the result.

FAQ

Related Strings functions

Open-source implementation

Unlike proprietary runtimes, every RunMat function is open-source. Read exactly how doc2sequence is executed, line by line, in Rust.

• View the source for doc2sequence in Rust on GitHub
• Learn how the RunMat runtime works
• Found a bug? Open an issue with a minimal reproduction.

About RunMat

RunMat is an open-source runtime that executes MATLAB-syntax code blazing on any GPU. It is licensed under the Apache 2.0 license.

• RunMat automatically optimizes your math for GPU execution on Apple, Nvidia, and AMD hardware. No code changes needed. Simulations that took hours now take minutes.
• Start running code in seconds. RunMat runs in the browser, on the desktop, or from the CLI. No license server, no IT ticket.

Getting started · Benchmarks · Pricing