v2.3.0

Added

• CSV exports create human readable folders and can output time information as datestrings instead of epochs.
• Added option for Parquet exports.
• Block Merging: Small pieces of new data written to AtriumDB now have the option to be merged into existing blocks, maintaining higher efficiency get_data calls.
• Data Transfer De-identification Files can now be reused even if the source dataset has added patients since creation.
• Unused TSC files can be detected and removed.
• get_device_patient_data now truncates its output to agree with given start/end time parameters.
• MariaDB SQL Handler now uses custom Connection Manager to leave open a single connection for all SDK uses.
• TSC Exports now, where possible, reuse blocks from the source SDK instead of re-encoding them, dramatically increasing transfer speeds.

Fixed

• Intervals of continuous data will now be properly bounded to the requested start/end times.
• Intervals wholly within other intervals will no longer cause interval data to be deleted.
• User entered parameters that will end up in the back end database are now force cast to Python native types to avoid known issues writing numpy objects to SQL/YAML.
• Blocks now concatenate correctly when reading a large sequential group of blocks in a single file.
• Intervals queried by patient are now properly bounded.
• Encounters with NULL end times (encounter ongoing) will be replaced with time.time_ns().

Removed

• The iterator number of samples per window can no longer be a floating point value (will be rounded to nearest int).