Reading/Writing#

Daft can read data from a variety of sources, and write data to many destinations.

Reading Data#

From Files#

DataFrames can be loaded from file(s) on some filesystem, commonly your local filesystem or a remote cloud object store such as AWS S3.

Additionally, Daft can read data from a variety of container file formats, including CSV, line-delimited JSON and Parquet.

Daft supports file paths to a single file, a directory of files, and wildcards. It also supports paths to remote object storage such as AWS S3.

import daft

# You can read a single CSV file from your local filesystem
df = daft.read_csv("path/to/file.csv")

# You can also read folders of CSV files, or include wildcards to select for patterns of file paths
df = daft.read_csv("path/to/*.csv")

# Other formats such as parquet and line-delimited JSON are also supported
df = daft.read_parquet("path/to/*.parquet")
df = daft.read_json("path/to/*.json")

# Remote filesystems such as AWS S3 are also supported, and can be specified with their protocols
df = daft.read_csv("s3://mybucket/path/to/*.csv")

To learn more about each of these constructors, as well as the options that they support, consult the API documentation on creating DataFrames from files.

From Data Catalogs#

If you use catalogs such as Apache Iceberg or Hive, you may wish to consult our user guide on integrations with Data Catalogs: Daft integration with Data Catalogs.

From File Paths#

Daft also provides an easy utility to create a DataFrame from globbing a path. You can use the daft.from_glob_path() method which will read a DataFrame of globbed filepaths.

df = daft.from_glob_path("s3://mybucket/path/to/images/*.jpeg")

# +----------+------+-----+
# | name     | size | ... |
# +----------+------+-----+
#   ...

This is especially useful for reading things such as a folder of images or documents into Daft. A common pattern is to then download data from these files into your DataFrame as bytes, using the .url.download() method.

From Memory#

For testing, or small datasets that fit in memory, you may also create DataFrames using Python lists and dictionaries.

# Create DataFrame using a dictionary of {column_name: list_of_values}
df = daft.from_pydict({"A": [1, 2, 3], "B": ["foo", "bar", "baz"]})

# Create DataFrame using a list of rows, where each row is a dictionary of {column_name: value}
df = daft.from_pylist([{"A": 1, "B": "foo"}, {"A": 2, "B": "bar"}, {"A": 3, "B": "baz"}])

To learn more, consult the API documentation on creating DataFrames from in-memory data structures.

Writing Data#

The df.write_*(…) methods are used to write DataFrames to files or other destinations.

# Write to various file formats in a local folder
df.write_csv("path/to/folder/")
df.write_parquet("path/to/folder/")

# Write DataFrame to a remote filesystem such as AWS S3
df.write_csv("s3://mybucket/path/")

Note that because Daft is a distributed DataFrame library, by default it will produce multiple files (one per partition) at your specified destination.