Skip to content

Fixed Overfitting with Dropout Rate, Weight Decay, and on the fly data augmentation. #18

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
cwsmith merged 5 commits into from
Nov 12, 2025
Merged

Fixed Overfitting with Dropout Rate, Weight Decay, and on the fly data augmentation. #18

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
df2dc03
Fixed overfitting by adding dropout and weight decay. Also added on t…
sridhs21 Nov 11, 2025
3f9364a
Updated README with flags.
sridhs21 Nov 11, 2025
a2fff6c
changed defaults in readme file
sridhs21 Nov 11, 2025
fa1ea6e
Made dropout_rate a required parameter in UNet.__init__
sridhs21 Nov 11, 2025
5f30b94
Changed default checkpoint interval from 10 to 100.
sridhs21 Nov 11, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 14 additions & 7 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,10 +85,12 @@ run it
The classifier supports several command line options for training configuration:

### Training Parameters
- `--learningRate`: Learning rate for training (default: 1e-4)
- `--batchSize`: Batch size for training (default: 8)
- `--epochs`: Number of training epochs (default: 100)
- `--minTrainingLoss`: Minimum reduction in training loss in orders of magnitude (default: 2, set to 0 to disable check)
- `--learningRate`: Learning rate for training (default: 1e-5)
- `--weightDecay`: Weight decay for L2 regularization (default: 5e-4)
- `--dropoutRate`: Dropout rate for regularization (default: 0.3)
- `--batchSize`: Batch size for training (default: 1)
- `--epochs`: Number of training epochs (default: 2000)
- `--minTrainingLoss`: Minimum reduction in training loss in orders of magnitude (default: 3, set to 0 to disable check)

### Data Configuration
- `--trainFrameFirst`: First frame number for training data (default: 1)
Expand All @@ -102,11 +104,13 @@ The classifier supports several command line options for training configuration:
- `--use-amp`: Enable automatic mixed precision training for faster training on modern GPUs
- `--amp-dtype`: Data type for mixed precision (`float16` or `bfloat16`, default: `bfloat16`)
- `--patience`: Patience for early stopping (default: 15 epochs)
- `--seed`: Random seed for reproducibility (default: None for non-deterministic)
- `--require-gpu`: Require GPU to be available, exit if not found

### Output and Monitoring
- `--plot`: Enable creation of figures showing ground truth and model-identified X-points
- `--plotDir`: Directory where figures are written (default: `./plots`)
- `--checkPointFrequency`: Number of epochs between model checkpoints (default: 10)
- `--checkPointFrequency`: Number of epochs between model checkpoints (default: 100)

### Performance Benchmarking
- `--benchmark`: Enable performance benchmarking (tracks timing, throughput, GPU memory)
Expand All @@ -118,18 +122,21 @@ The classifier supports several command line options for training configuration:

### Example with Advanced Options

For faster training with mixed precision and early stopping:

For training with custom regularization and reproducibility:
```bash
python -u ${rcRoot}/reconClassifier/XPointMLTest.py \
--paramFile=/path/to/params.txt \
--xptCacheDir=/path/to/cache \
--epochs 200 \
--learningRate 1e-4 \
--weightDecay 1e-3 \
--dropoutRate 0.3 \
--batchSize 16 \
--use-amp \
--amp-dtype bfloat16 \
--patience 20 \
--seed 42 \
--require-gpu \
--plot \
--trainFrameLast 100 \
--validationFrameLast 120
Expand Down
Loading