Tabalyst
DOCS
Report guide
Commands, options and outputs of Tabalyst Report.
Tabalyst for data professionals
Understand the dataset before analysis or modeling.
Before the first chart or the first model, see what the file holds: how complete it is, what each column contains, where values are rare or inconsistent. Tabalyst Report reads the whole file on your machine and gives you an interactive report to explore.
tabalyst report insurance-customers.csv
Analyzed 3,000 rows and 34 columns.
Report: insurance-customers.html
tabalyst sample insurance-customers.csv --sample-method random --rows 1000 --seed 42
Sample created successfully.
Source rows: 3,000
Sample rows: 1,000
Real output of tabalyst 0.6.0, paths shortened. Some lines of the sample summary are left out.
Your angle
What you want to know first
- 01 / OVERVIEW
Completeness and shape first
Rows, columns, missing cells, duplicate rows and the types found, on one screen, before you open a notebook. In the example report, 15 of 34 columns carry an observation.
- 02 / DISTRIBUTIONS
The values, not only the types
Range, mean and median of numeric columns, string lengths, the frequency of each value in a column with few distinct values, and representative examples. Odd spellings and rare values stand out.
- 03 / DOUBT
Ambiguity is shown, not hidden
An ambiguous date stays ambiguous: the report shows the evidence of the column and does not choose for you. Missing values are counted by kind: absent, null, empty, blank, or a marker such as N/A.
Workflow
From a file to a dataset you understand
The report is one HTML file. Open it in a browser, work through its sections, and keep the JSON profile if you want the same facts in code.
- 1 · Install
pip install tabalyst# Also updates it: Tabalyst changes often
- 2 · Report a file
tabalyst report data.csv# data.html to explore, data.json for code, beside the source
- 3 · Work on a sample
tabalyst sample data.csv --sample-method random --rows 1000 --seed 42# data.sample.csv: a smaller CSV, the source is never modified
import tabalyst # The report and the profile of a fileprofile = tabalyst.analyze("data.csv", "reports/data.html") # A reproducible sample, to try ideas on a smaller filesample = tabalyst.sample_csv("data.csv", method="random", rows=1000, seed=42)print(sample.output)
Sampling methods are first, last, random and stratified, which approximately preserves the distribution of a field you choose. Sampling reads CSV files only.
What each section of the report answers
- Overview: the signals of the whole file.
- Columns: the type, the missing values and the issues of every column.
- Numeric, Strings: ranges, means and medians; the length profile of text columns.
- Dates: the formats found, and which values are ambiguous or invalid.
- Detectors: what was recognized, such as email addresses, phone numbers and postal codes.
- Sample, Settings: the raw values as they are in the file, and every rule used.
Proof
Explore a real report
TabalystReport
Explore the report
Report
Analysis
Data
The demo shows screenshots of a report on 3,000 synthetic insurance customers. The full report and its profile are the files Tabalyst Report wrote.
Scope, not hype
What a report does not tell you
- A report counts what is in the file. A column that is mostly empty can be normal or a problem: that judgment is yours.
- Tabalyst Report does not clean the file, fill gaps, model the data or test hypotheses.
- The report and the profile may contain values of the source, even though sensitive fields such as email addresses are masked by default. Share them accordingly.
- Sampling reads CSV files only.
The known limitations ↗ list the rest.
Next step
Where to go next
Tabalyst Report summarizes and profiles a file. Tabalyst Explore, which will filter and investigate it interactively, is still to come.
Coming soon Tabalyst Explore: guided investigation of your data
Use Tabalyst Report todayDOCS
Tabalyst Sample
Random, stratified, first and last samples of a CSV file.
Your next dataset
See the data before you model it.
Install Tabalyst, run it on your file and open the report.