CSV & Excel Data Profiler (EDA Report)

About the CSV & Excel Data Profiler (EDA Report)

Open a CSV or Excel file and get an exploratory data analysis of it in one step: for every column its inferred type (numbers, dates, text, yes/no, e-mails, phone numbers, web addresses, UUIDs), missing values, distinct values, minimum, maximum, mean, median and standard deviation, the most common values, text lengths, value patterns and a small histogram.

The report also warns about what usually breaks an import or an analysis: duplicate rows, empty rows and columns, constant columns, columns that mix types, dates that could be read two ways, codes with leading zeros, numbers written with currency signs, likely ID columns and columns that may hold personal data. It writes a printable report of the whole profile (a web page to print or save as PDF), a data dictionary (Markdown or CSV) and the inferred schema as SQL CREATE TABLE, JSON Schema, a Frictionless Table Schema or a BigQuery JSON schema. The file is read and profiled in your browser; nothing is uploaded.

Common uses

• Understanding a file someone sent before working with it: what each column holds and how complete it is.
• Checking data quality before an import, a report or a machine-learning experiment.
• Writing the CREATE TABLE statement or the BigQuery schema for loading a CSV into a database.
• Documenting a dataset with a data dictionary for a team, a client or a data catalogue.
• Finding duplicate rows, empty columns and inconsistent values at a glance.

How column types are decided

Every filled value is classified as a whole number, a decimal number (including values written with digit grouping, currency or percent signs), a date, a date and time, a time of day, an e-mail address, a web address, a UUID, a phone number or text. A column takes the type of at least 80 % of its values — whole numbers count as decimals and dates as dates and times when the two are mixed — and the rest are listed as mixed values with examples. A column whose values are all one of the pairs yes/no, true/false, y/n or t/f is yes/no.

Two rules keep data safe: whole numbers with a leading zero (postal codes, account numbers) stay text, and dates such as 04/10/2026 are read in the order the column itself proves (any day above 12 decides it) or, when nothing decides it, in the order you choose. Columns of more than 100,000 values are classified on an even sample of 100,000; every count and statistic uses all values.

What a column card shows

• Filled, missing and distinct values; a column whose values are all filled and different is a likely ID or key.
• Numbers: minimum, maximum, mean, median, quartiles (Hyndman–Fan type 7, as in Excel’s QUARTILE.INC), standard deviation (n − 1), sum, zeros, negatives and outliers beyond 1.5 × IQR.
• Dates: the earliest and latest, the span in days and how they are written.
• Top values with counts, text lengths and the three most common patterns, where letters become A or a and digits 9 (“AB-1234” is “AA-9999”).
• A histogram: number ranges about as wide as the Freedman–Diaconis rule suggests (2 × IQR × n^−1/3), dates by year, month or day, and the most common values of text columns.

Schemas and the data dictionary

• SQL CREATE TABLE for PostgreSQL, MySQL/MariaDB, SQLite or SQL Server: integer, decimal (with the precision and scale the values need, currency and percent signs aside), boolean, date, timestamp, time, UUID and text types — whole numbers longer than 18 digits, which a 64-bit integer cannot hold, as exact decimals; NOT NULL for columns without missing values; a primary key for a likely ID column named like id, key or code; and a comment where a column needs converting first (dates not written YYYY-MM-DD, numbers with currency signs).
• JSON Schema (draft 2020-12) for the rows as an array of objects, with formats for dates, date-times written the RFC 3339 way, e-mails, web addresses and UUIDs, and optionally the observed ranges and value lists as constraints.
• A Frictionless Table Schema with types, date patterns such as %d/%m/%Y (or %y for two-digit years), digit grouping (groupChar) and signs (bareNumber) of numbers, required and unique constraints and the missing-value markers found.
• A BigQuery JSON schema (schema format) with valid column names, INT64, NUMERIC, FLOAT64, BOOL, DATE, DATETIME and TIMESTAMP types and REQUIRED or NULLABLE modes.

When a column name repeats, the SQL and JSON schemas number the repeats (Amount, Amount_2) and say so in a comment, because a table cannot have two columns of one name. The data dictionary lists every column with its type, missing values, distinct values, range, most common values and notes. Schemas are inferred from the values in the file: check them before using them, for example widen a text column that will hold longer values later.

Limitations

• Up to 500,000 rows are profiled (text files up to 100 MB, workbooks up to 50 MB, as far as your device’s memory allows); more rows are left out with a note.
• Types are inferred from the values, not from meaning: a column of years is a column of whole numbers, and a phone number written without spaces or a + is a number.
• Excel files are read as their shown values; formulas count as their last saved results.
• The printable report is an HTML file, not a PDF: open it in a browser and print it or save it as PDF from the print dialog.

Privacy

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot.