Dataset Profiler overview
Use the Dataset Profiler to inspect an uploaded dataset without modifying the source file. It reports row and column counts, duplicate rows, mixed-type columns, schema fields, statistics, and data-quality warnings. Processing and calculations run in the browser.
How to profile a dataset
- Select a CSV, TSV, TXT, JSON, JSONL, NDJSON, XLSX, or XLS file with browse files, or drop it onto the upload area. TXT files are read as comma-delimited data; NDJSON files are read one JSON record per line.
- Review the summary for Rows, Columns, Duplicate rows, and Mixed-type columns.
- Read the schema table, which shows Column, Type, Null rate, Blank rate, Unique, Mean, and Warnings.
- Click Details for one column to open its drill-down with type counts, available statistics, warnings, and top values.
- Choose Export JSON, Export CSV, or Export HTML.
What the profile shows
Each non-missing value is classified as integer, number, boolean, date, or text. The most common type becomes the column type. A column is marked mixed when more than 2% of its present values differ from that dominant type.
Null rate counts null or undefined values and configured missing tokens such as N/A, null, and -. Blank rate counts present empty strings. Unique values are counted among present, non-missing values.
Duplicate detection compares normalized row values. It trims strings, treats recognized missing values the same way, and counts extra occurrences after the first row. The top values list contains up to 10 values.
Numeric columns can show min, max, mean, median, standard deviation, and P10, P25, P75, and P90 values. Date columns show earliest and latest values. Text-like columns show minimum, maximum, and average length. Warnings use heuristics such as a per-column null rate above 20%, dataset-level reporting for columns above 50% null, blank strings, mixed types, inconsistent casing, constant columns, and likely identifier columns.
Exporting the results
Export JSON contains the broadest profile object, including column type counts, statistics, top values, and warnings. Export CSV contains schema-level fields: column, type, present, nulls, null_rate, blanks, blank_rate, unique, unique_rate, min, max, mean, median, stddev, and warnings. Export HTML is a schema summary with the displayed Mean field. The CSV and HTML exports omit some drill-down fields, including type counts, top values, quantiles, and text statistics.
Input requirements and limitations
- CSV and TSV files use the first row as headers. If a header is empty, the tool can generate a name such as
Column 1. - Spreadsheet profiling reads only the first worksheet.
- A JSON file can contain a top-level array or an object with a first qualifying nested array. For useful profiling, provide records as objects. JSONL and NDJSON files are read one JSON record per line.
- Invalid JSON, JSONL, NDJSON, or spreadsheet data can show a read error. An input that produces no rows shows a no-rows error. The local CSV/TSV parser does not establish that every malformed delimited structure will be rejected.
- The implementation does not document a maximum file size, row count, column count, or processing-time limit. Browser compatibility is also not documented.
What this tool does not do
The profiler summarizes schema, completeness, types, and distributions. It does not detect statistical outliers, compute correlations, or flag time-series anomalies. It also does not rewrite the source file or alter its values.
Frequently Asked Questions
Does this tool modify my data?
No. It only reads your file to compute profiling statistics for display and export. It never rewrites your source file or alters any values.
How is the data type of a column decided?
Each non-missing value is classified individually, and the most common type in a column becomes that column's reported type. A column is flagged as mixed when more than 2% of its present values differ from that dominant type.
What counts as a duplicate row?
A row is a duplicate when its normalized values match another row. The tool trims strings, normalizes recognized missing values, and counts extra occurrences beyond the first.
Does this tool find outliers, correlations, or time-series anomalies?
No. This tool focuses on schema, completeness, type inference, and distribution summaries. Dedicated outlier detection, correlation-matrix, and time-series anomaly tools handle those focused analytical questions separately.