🚀 RE: Active-Negative-Loss-Functions

🎨 Original Work

Explore the groundbreaking work at Virusdoll's GitHub repository.

🛠️ Requirements

• Python >= 3.9
• Torch >= 1.12.1
• Torchvision >= 0.13.1
• Numpy >= 1.23.1

🛠️ Changes Made

We've wielded our coding wand to enhance utils.py:

• ✨ Added conditions for loss functions nce and nnce.
• 🛠️ Squashed the pesky MAE error.

📊 Reproduced Results

Marvel at our scientific endeavors in the "experiment" folder.

Tables Regenerated:

📈 Reproduced Graphs

Behold the visual symphony we've composed:

To summon these magical graphs:

1. 🧙‍♂️ Navigate to the desired folder.
2. 🪄 Cast the spell by executing Graphgen.py.

For example, to conjure the graph of ANL_CE on CIFAR10:

• 🌌 Venture into the anl_ce_cifar10_graph realm.
• 🌟 Invoke the Python file.

Additionally, we offer tensorboard support (with credit to the original authors):

1. 🚀 Launch tensorboard with tensorboard --logdir=runs\cifar10\sym\anl_ce.
2. 🌈 Witness the creation of wondrous graphs.

🙏 Acknowledgements

We extend our deepest gratitude to the visionary creator, Virusdoll. Your brilliance lights our path! 💫