Skeem

About

Skeem infers SQL DDL statements from tabular data.

Skeem is, amongst others, based on the excellent ddlgenerator, frictionless, fsspec, pandas, ScipPy, SQLAlchemy and xarray packages, and can be used both as a standalone program, and as a library.

Supported input data:

• Apache Parquet
• CSV
• Google Sheets
• GRIB
• InfluxDB line protocol
• JSON
• NetCDF
• NDJSON (formerly LDJSON) aka. JSON Lines, see also JSON streaming
• Office Open XML Workbook (Microsoft Excel)
• OpenDocument Spreadsheet (LibreOffice)

Supported input sources:

• Amazon S3
• File system
• GitHub
• Google Cloud Storage
• HTTP

Please note that Skeem is beta-quality software, and a work in progress. Contributions of all kinds are very welcome, in order to make it more solid. Breaking changes should be expected until a 1.0 release, so version pinning is recommended, especially when you use it as a library.

Synopsis

skeem infer-ddl --dialect=postgresql data.ndjson

CREATE TABLE "data" (
    "id" SERIAL NOT NULL,
    "name" TEXT NOT NULL,
    "date" TIMESTAMP WITHOUT TIME ZONE,
    "fruits" TEXT NOT NULL,
    "price" DECIMAL(2, 2) NOT NULL,
    PRIMARY KEY ("id")
);

Quickstart

If you are in a hurry, and want to run Skeem without any installation, just use the OCI image on Podman or Docker.

docker run --rm ghcr.io/daq-tools/skeem-standard \
    skeem infer-ddl --dialect=postgresql \
    https://github.com/daq-tools/skeem/raw/main/tests/testdata/basic.ndjson

Setup

Install Skeem from PyPI.

pip install skeem

Install Skeem with support for additional data formats like NetCDF.

pip install 'skeem[scientific]'

Usage

This section outlines some example invocations of Skeem, both on the command line, and per library use. Other than the resources available from the web, testing data can be acquired from the repository's testdata folder.

Command line use

Help

skeem info
skeem --help
skeem infer-ddl --help

Read from files

# NDJSON, Parquet, and InfluxDB line protocol (ILP) formats.
skeem infer-ddl --dialect=postgresql data.ndjson
skeem infer-ddl --dialect=postgresql data.parquet
skeem infer-ddl --dialect=postgresql data.lp

# CSV, JSON, ODS, and XLSX formats.
skeem infer-ddl --dialect=postgresql data.csv
skeem infer-ddl --dialect=postgresql data.json
skeem infer-ddl --dialect=postgresql data.ods
skeem infer-ddl --dialect=postgresql data.xlsx
skeem infer-ddl --dialect=postgresql data.xlsx --address="Sheet2"

Read from URLs

# CSV, NDJSON, XLSX
skeem infer-ddl --dialect=postgresql https://github.com/daq-tools/skeem/raw/main/tests/testdata/basic.csv
skeem infer-ddl --dialect=postgresql https://github.com/daq-tools/skeem/raw/main/tests/testdata/basic.ndjson
skeem infer-ddl --dialect=postgresql https://github.com/daq-tools/skeem/raw/main/tests/testdata/basic.xlsx --address="Sheet2"

# Google Sheets: Address first sheet, and specific sheet of workbook.
skeem infer-ddl --dialect=postgresql --table-name=foo https://docs.google.com/spreadsheets/d/1ExyrawjlyksbC6DOM6nLolJDbU8qiRrrhxSuxf5ScB0/view
skeem infer-ddl --dialect=postgresql --table-name=foo https://docs.google.com/spreadsheets/d/1ExyrawjlyksbC6DOM6nLolJDbU8qiRrrhxSuxf5ScB0/view#gid=883324548

# InfluxDB line protocol (ILP)
skeem infer-ddl --dialect=postgresql https://github.com/influxdata/influxdb2-sample-data/raw/master/air-sensor-data/air-sensor-data.lp

# Compressed files in gzip format
skeem --verbose infer-ddl --dialect=crate --content-type=ndjson https://s3.amazonaws.com/crate.sampledata/nyc.yellowcab/yc.2019.07.gz

# CSV on S3
skeem --verbose infer-ddl --dialect=postgresql s3://noaa-ghcn-pds/csv/by_year/2022.csv

# CSV on Google Cloud Storage
skeem --verbose infer-ddl --dialect=postgresql gs://tinybird-assets/datasets/nations.csv
skeem --verbose infer-ddl --dialect=postgresql gs://tinybird-assets/datasets/medals1.csv

# CSV on GitHub
skeem --verbose infer-ddl --dialect=postgresql github://daq-tools:skeem@/tests/testdata/basic.csv

# GRIB2, NetCDF
skeem infer-ddl --dialect=postgresql https://github.com/earthobservations/testdata/raw/main/opendata.dwd.de/weather/nwp/icon/grib/18/t/icon-global_regular-lat-lon_air-temperature_level-90.grib2
skeem infer-ddl --dialect=postgresql https://www.unidata.ucar.edu/software/netcdf/examples/sresa1b_ncar_ccsm3-example.nc
skeem infer-ddl --dialect=postgresql https://www.unidata.ucar.edu/software/netcdf/examples/WMI_Lear.nc

OCI

OCI images are available on the GitHub Container Registry (GHCR). In order to run them on Podman or Docker, invoke:

docker run --rm ghcr.io/daq-tools/skeem-standard \
    skeem infer-ddl --dialect=postgresql \
    https://github.com/daq-tools/skeem/raw/main/tests/testdata/basic.csv

If you want to work with files on your filesystem, you will need to either mount the working directory into the container using the --volume option, or use the --interactive option to consume STDIN, like:

docker run --rm --volume=$(pwd):/data ghcr.io/daq-tools/skeem-standard \
    skeem infer-ddl --dialect=postgresql /data/basic.ndjson

docker run --rm --interactive ghcr.io/daq-tools/skeem-standard \
    skeem infer-ddl --dialect=postgresql --content-type=ndjson - < basic.ndjson

In order to always run the latest nightly development version, and to use a shortcut for that, this section outlines how to use an alias for skeem, and a variable for storing the input URL. It may be useful to save a few keystrokes on subsequent invocations.

docker pull ghcr.io/daq-tools/skeem-standard:nightly
alias skeem="docker run --rm --interactive ghcr.io/daq-tools/skeem-standard:nightly skeem"
URL=https://github.com/daq-tools/skeem/raw/main/tests/testdata/basic.ndjson

skeem infer-ddl --dialect=postgresql $URL

More

Use a different backend (default: ddlgen):

skeem infer-ddl --dialect=postgresql --backend=frictionless data.ndjson

Reading data from STDIN needs to obtain both the table name and content type separately:

skeem infer-ddl --dialect=crate --table-name=foo --content-type=ndjson - < data.ndjson

Reading data from STDIN also works like this, if you prefer to use pipes:

cat data.ndjson | skeem infer-ddl --dialect=crate --table-name=foo --content-type=ndjson -

Library use

import io
from skeem.core import SchemaGenerator
from skeem.model import Resource, SqlTarget

INDATA = io.StringIO(
    """
    {"id":1,"name":"foo","date":"2014-10-31 09:22:56","fruits":"apple,banana","price":0.42}
    {"id":2,"name":"bar","date":null,"fruits":"pear","price":0.84}
    """
)

sg = SchemaGenerator(
    resource=Resource(data=INDATA, content_type="ndjson"),
    target=SqlTarget(dialect="crate", table_name="testdrive"),
)

print(sg.to_sql_ddl().pretty)

CREATE TABLE "testdrive" (
    "id" INT NOT NULL,
    "name" STRING NOT NULL,
    "date" TIMESTAMP,
    "fruits" STRING NOT NULL,
    "price" DOUBLE NOT NULL,
    PRIMARY KEY ("id")
);

Development

For installing the project from source, please follow the development documentation.

Project information

Credits

• Catherine Devlin for ddlgenerator and data_dispenser.
• Mike Bayer for SQLAlchemy.
• Paul Walsh and Evgeny Karev for frictionless.
• Wes McKinney for pandas.
• All other countless contributors and authors of excellent Python packages, Python itself, and turtles all the way down.

Prior art

We are maintaining a list of other projects with the same or similar goals like Skeem.

Etymology

The program was about to be called Eskema, but it turned out that there is already another Eskema out there. So, it has been renamed to Skeem, which is Estonian, and means "schema", "outline", or "(to) plan".