active 2025-05-03 → 2025-05-20 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 18 days from 2025-05-03 to 2025-05-20. Pushes: 20 total, peak 8 in a day. Pull requests: 6 total, peak 4 in a day. Issues: 16 total, peak 8 in a day. Comments: 0 total, peak 0 in a day. Stars: 0 total, peak 0 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| valterstra | 22 | 14 | 2 | 0 |
| wiande | 12 | 5 | 4 | 0 |
| Oscar2343 | 8 | 1 | 0 | 0 |
Recent activity
Latest issues, pull requests and releases
- Pull request#11valterstra2025-05-20 09:01
- Pull request#11valterstra2025-05-20 09:00
- Issue#7wiande2025-05-10 20:58Replace the fully-connected layer after the last convolution layer with a global average pooling layer. Does this affect the amount of regularization needed and/or the final test performance on the ?
- Issue#4wiande2025-05-10 20:48Apply slightly more sophisticated data-augmentation such as label-smoothing and/or cut-out (ie masking out random rectangles in an image)
- Pull request#10wiande2025-05-10 20:47
- Pull request#10wiande2025-05-10 20:47
- Issue#6valterstra2025-05-10 20:03Perform the down-sampling by applying the second convolution in each VGG block with stride 2 and then remove the max-pooling layer
- Issue#5valterstra2025-05-10 19:21Try different learning rate schedulers such as learning rate warm-up + cosine annealing, step decay or cosine annealing with re-starts and see how is helps/affects training.
- Issue#3valterstra2025-05-10 14:39Data-augmentation: horizontal flipping + x− and y− translation shifts.
- Issue#9valterstra2025-05-10 14:39Then you should train the network with multiple regularization strategies combined to increase performance plus the addition of batch normalization
- Issue#9valterstra2025-05-10 11:51Then you should train the network with multiple regularization strategies combined to increase performance plus the addition of batch normalization
- Issue#1valterstra2025-05-10 11:50Dropout regularly applied throughout the network.
- Pull request#8wiande2025-05-10 11:31
- Pull request#8wiande2025-05-10 11:27
- Issue#2wiande2025-05-07 19:03Weight decay a.k.a. L2 regularization.
- Issue#7Oscar23432025-05-03 09:18Replace the fully-connected layer after the last convolution layer with a global average pooling layer. Does this affect the amount of regularization needed and/or the final test performance on the ?
- Issue#6Oscar23432025-05-03 09:18Perform the down-sampling by applying the second convolution in each VGG block with stride 2 and then remove the max-pooling layer
- Issue#5Oscar23432025-05-03 09:18Try different learning rate schedulers such as learning rate warm-up + cosine annealing, step decay or cosine annealing with re-starts and see how is helps/affects training.
- Issue#4Oscar23432025-05-03 09:17Apply slightly more sophisticated data-augmentation such as label-smoothing and/or cut-out (ie masking out random rectangles in an image)
- Issue#3Oscar23432025-05-03 09:17Data-augmentation: horizontal flipping + x− and y− translation shifts.
- Issue#2Oscar23432025-05-03 09:17Weight decay a.k.a. L2 regularization.
- Issue#1Oscar23432025-05-03 09:17Dropout regularly applied throughout the network.
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 0 stars here means stars gained during the window, not the repo's star count.