Change Log¶
All notable changes to PyGOP will be documented here.
[0.2.3] 2020-09-15¶
Added¶
- recommendation for tensorflow-1.14.0
Fixed¶
- fixing keras version to 2.2.4 in requirements and setup file
[0.2.2] 2019-04-24¶
Added¶
- Test Suite: unit test, PEP8 verification
- Continuous Integration (Travis)
- Mnist Example
- Instructions on Contributions
- Improved Documentation
- requirements.txt
- Functionality to specify optimizer’s parameters
Fixed¶
- learning rate update in utility/gop_utils.network_trainer()
- improve coding style according to PEP8
[0.2.1] 2018-12-29¶
Added¶
- new wheel for version 0.2.1 in releases
Fixed¶
- move the tmp files removal step from progressive_learn to the end of fit() in all models (hemlgop.py, homlgop.py, hemlrn.py homlrn.py, pop.py, popfast.py, popmemo.py, popmemh.py). This fixes the bug that removes train_states.pickle before finetuning.
- change file opening option from ‘r’ to ‘rb’ in utility/misc.initialize_states() to read train_states.pickle in Python3
[0.2.0] 2018-12-17¶
Added¶
- utility/block_update.py
- utility/calculate_memory.py
- Added CHANGELOG.md to keep track of major changes
- Added releases directory to keep track of different wheel versions.
- Functionalities to spawn new process when calculating memory block (utility/gop_utils.calculate_memory_block_standalone() and utility/calculate_memory.py). This prevents potential OOM errors when tensorflow-gpu does not release memory right after the block finishes.
- Functionalities to spawn new process when finetuning some blocks (utility/gop_utils.block_update_standalone() and utility/block_update.py). This also prevents potential OOM errors mentioned above.
Fixed¶
- utility/misc.check_model_parameters()
- models/_model.print_performance()