Skip to content

Bodo 2021.5 Release (Date: 5/19/2021)

This release includes many new features, optimizations, bug fixes and usability improvements. Overall, 70 code patches were merged since the last release.

New Features and Improvements

  • Bodo is updated to use Arrow 4.0 (latest)

  • Connectors:

    • Improved performance of pd.read_parquet significantly for large multi-file datasets by optimizing Parquet metadata collection
    • Bodo nows reads only the first few rows from a Parquet dataset if the program only requires df.head(n) and/or df.shape. This helps with exploring large datasets without the need for a large cluster to load the full data in memory.
  • Visualization: Bodo now supports calling many Matplotlib plotting functions directly from JIT code. See the "Data Visualization" section of our documentation for more details. The current support gathers the data into one process but this will be avoided in future releases.

  • Improved compilation time for dataframe functions

  • Improved the performance and scalability of groupby.nunique

  • Many improvements to error checking and reporting

  • Bodo now avoids printing empty slices of distributed data to make print output easier to read.

  • Pandas coverage:

    • Support for
    • Support for memory_usage() for DataFrame and Series
    • Support for nbytes for array and Index types
    • Support for df.describe() with datetime data (assumes datetime_is_numeric=True)
    • Support for groupby.value_counts()
    • Support for pd.NamedAgg with nunique in groupby
    • Initial support for CategoricalIndex type and categorical keys in groupby
    • Support for groupby idxmin and idxmax with nullable Integer and Boolean arrays
    • Support for timedelta64 in Groupby.agg
    • Support for bins and other optional arguments in Series.value_counts()
    • Support for df.dtypes
    • Support passing df.dtypes to df.astype(), for example: df1.astype(df2.dtypes)
    • Support for boolean pd.Index
    • Support for Series.sort_index()
    • Support for Timestamp.day_name() and Series.dt.day_name()
    • Support for Series.quantile() with datetime
    • Support for passing list of quantile values to Series.quantile()
    • Support for Series.to_frame()
    • Support for sum() method of Boolean Arrays
    • Initial support for MultiIndex.from_product
    • String array comparison returns a Pandas nullable boolean array instead of a Numpy boolean array