v1.5.3

Improvements:

• Increased CHUNK size as per miniz instructions
• Added tests for very large arrays in NPZ files
• Added some CI tests to catch issues across platforms
• Removed the internal IO streams in favor of just using stringstream
• NPZs can now be read from and written to memory

Bugfixes:

• Fixed an issue where very large arrays in NPZ files would throw an error
• Fixed a bug with mac builds due to deprecated APIs

Version 1.5.2

Removing using namespace std to simplify library use

Version 1.5.1

Improvements:

• CMake build now uses the highest compiler warning/error setting

Bugfixes:

• Fixed some bugs exposed by heightened compiler warnings

Version 1.5.0

Improvements:

• Added a keys member to inpzstream so it is possible to query the keys of the tensors

Version 1.4.1

Bug fixes:

• Fixed a bug with integer shifting

Version v1.4.0

Improvements:

• Further minor CMake changes to improve ease of use
• NPZ streams now have is_open methods to check for successful file opening
• Minor code style changes

Bug fixes:

• NPZ files will now correctly handle PKZIP versions after 2.0, both for reading and writing

Version 1.3.1

Improvements:

• Updated CMake integration to make the library easier to use via FetchContent

Version 1.3.0 - Unicode support

New Features:

• Support for Unicode string tensors (npy type 'U')

Breaking change:

• CopyFrom interface for C# Tensors has been changed to use *Buffer objects

Version 1.2.2

Improvements:

• Bug fix for a missing comma on 1d shape

Version 1.2.1

Improvements:

• Bug fix for scalar tensor reading
• Bug fix with memstream buffer size at initialization
• ".npy" will be added to tensor names in NPZ writing if not already present