matlab-import-export-data

# MATLAB Data Import/Export

Guidance for MATLAB data I/O — correct patterns for delimiters, locales, format-specific quirks, and common error messages.

## When to Use

- Reading or writing CSV, Excel, Parquet, or JSON files in MATLAB with `readtable`/`writetable`/`detectImportOptions`
- Troubleshooting data I/O errors (misleading messages, "file not found" variants)
- Reading data from URLs or authenticated REST API endpoints
- Importing non-English locale data (European decimals, semicolons)
- Validating imported data for silent corruption (NaN, 65535, type widening)
- Reading compressed files or JSON with non-identifier keys
- Reading or writing text files (use `readlines`/`writelines`, not `fopen`/`fgetl`/`fprintf`)
- Reading or writing XML files (use MAXP provider, not legacy JAXP)

## When NOT to Use

- Large files that may not fit in memory, or choosing between tall arrays, datastores, and parallel workflows (use `matlab-choose-big-data-solution` skill)
- Database access via ODBC/JDBC — reading, writing, or querying relational databases (use `matlab-use-database` skill)
- SQL-based queries on large CSV/Parquet/JSON files for reduction before analysis (use `matlab-use-duckdb` skill)
- Vehicle data from MDF/MF4/BLF/ASC log files or CAN/LIN bus decoding (use `matlab-import-export-vehicle-data` skill)
- Vehicle network communication setup with CAN/CAN FD/J1939 (use `matlab-use-vehicle-network` skill)
- Tracking data import for sensor fusion workflows (use `matlab-import-tracking-data` skill)
- Medical image data — DICOM, NIfTI, or Analyze formats (use `matlab-read-medical-data` skill)
- Market or financial data feeds (use `matlab-access-datafeed` skill)
- Simulink data logging or signal I/O (use Simulink-specific workflows)
- Image or audio file I/O (`imread`, `audioread` — different domain)
- Streaming or real-time data acquisition (use Data Acquisition Toolbox)
- File system operations, path manipulation, or folder traversal

## General principles

- **Validate after import.** Append 2-3 assertion-style checks that verify the imported data matches expectations.
- **Always set `TextType="string"` for text and Excel imports** — the `string` type is more efficient and easier to work with than char or cell arrays of character vectors. This only needs to be set for delimited text and spreadsheet formats; XML, JSON, and other formats already return strings by default.
- **Specify `FileType` when reading from URLs that lack a recognizable extension** — MIME type detection handles some cases, but API endpoints and non-standard URLs still need explicit format (see Topic 4). When using `detectImportOptions` with `readtable`, pass `FileType` on `detectImportOptions` — `readtable` does not accept `FileType` when an import options object is provided.
- **Handle missing value placeholders at read time** — use `TreatAsMissing` on `readtable` instead of calling `standardizeMissing` after import.
- **Read directly from compressed files** — `readtable`, `readmatrix`, `readtimetable`, and other read functions accept ZIP, GZ, and TAR file paths directly without manual extraction (R2025a+).
- **European CSVs use `;` as delimiter because `,` is the decimal separator** — set both `Delimiter` and `DecimalSeparator` when contextual cues suggest European-format data.

## Topics

### 1. Import Function Selection (Delimiter & Locale Handling)

When contextual cues suggest European-format data (German/French/Italian offices, semicolon-delimited files, column names in a European language), proactively set `Delimiter`, `DecimalSeparator`, and `Encoding` (UTF-8 for umlauts/accents):

```matlab
opts = detectImportOptions("messdaten.csv", ...
    "Delimiter", ";", "DecimalSeparator", ",", "Encoding", "UTF-8");
T = readtable("messdaten.csv", opts);
```

For files containing path-like data (`/data/exp_01/run_003/results.mat`), explicitly set the actual delimiter — detection may pick `/` from the path column:

```matlab
opts = detectImportOptions("fileList.csv");
opts.Delimiter = ",";
T = readtable("fileList.csv", opts);
```

For numeric data with embedded unit suffixes (e.g., `6.53e+001dB`, `-9.00e+001°`), use `TrimNonNumeric` (R2022a+) to strip non-numeric characters instead of `textscan` or `regexp`:

```matlab
T = readtable("circuit_output.txt", "Delimiter", {"\t", ","}, ...
    "NumHeaderLines", 1, "TrimNonNumeric", true);
```

`TrimNonNumeric` can be passed directly to `readtable` as a name-value pair, or set per-variable via `setvaropts(opts, vars, "TrimNonNumeric", true)` when only specific columns have suffixes.

---

### 2. Error Message Interpretation

MATLAB I/O error messages can be broad, pointing to a general category rather than the specific issue:

| Error Message | Likely Actual Cause | Recovery |
|--------------|-------------------|----------|
| `"Entry may be password-protected or encrypted"` | Disk space insufficient in temp directory for unzip (observed in R2020a–R2023b) | Check available space with `tempdir`; free space or redirect temp |
| `"Unrecognized file extension"` | URL lacks a recognizable file extension | Specify `FileType` name-value pair explicitly (see Topic 4) |

---

### 3. Import Validation & Data Fidelity

After importing from Excel or Parquet, check for silent data corruption:

- **Excel `Inf` → 65535**: Both `Inf` and `-Inf` are written as 65535. Values of exactly 65535 that seem physically implausible likely represent Inf.
- **Excel complex → NaN**: Excel cannot store complex numbers. An entirely NaN column from a spreadsheet may contain complex data in the source.
- **Parquet integer columns with nulls → silent type promotion**: When any integer column (int8/16/32/64, uint8/16/32/64) contains null values, `parquetread` promotes it to `double` (MATLAB integer types have no missing representation). For int64/uint64, values above 2^53 silently lose precision. Detect by comparing `parquetinfo` schema against `class(T.col)`. Workaround: use `parquetDatastore` with `ReadSize="fil