Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: pytorch
|
| 3 |
+
tags:
|
| 4 |
+
- vae
|
| 5 |
+
- genomics
|
| 6 |
+
- genome-minimization
|
| 7 |
+
- e-coli
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# Genome Minimizer 2 — V3
|
| 11 |
+
|
| 12 |
+
VAE model for generating minimal *E. coli* genomes, trained with the **v3** configuration.
|
| 13 |
+
|
| 14 |
+
## Model Details
|
| 15 |
+
|
| 16 |
+
| | |
|
| 17 |
+
|---|---|
|
| 18 |
+
| **Architecture** | VAE: 55,039 → 512 → 32 (latent) |
|
| 19 |
+
| **Parameters** | 57,521,983 |
|
| 20 |
+
| **Loss** | Reconstruction + KL divergence (cosine) + Weighted gene abundance + L1 regularization |
|
| 21 |
+
| **Epochs trained** | 2363 |
|
| 22 |
+
| **Test F1** | 0.9121 |
|
| 23 |
+
| **Test Accuracy** | 0.9861 |
|
| 24 |
+
|
| 25 |
+
## Training Configuration
|
| 26 |
+
|
| 27 |
+
| Parameter | Value |
|
| 28 |
+
|---|---|
|
| 29 |
+
| Hidden dim | 512 |
|
| 30 |
+
| Latent dim | 32 |
|
| 31 |
+
| Learning rate | 0.001 |
|
| 32 |
+
| Batch size | 32 |
|
| 33 |
+
| Beta range | 0.1 → 1.0 |
|
| 34 |
+
| Gamma range | 2.0 → 0.1 |
|
| 35 |
+
| L1 lambda | 0.01 |
|
| 36 |
+
| Weight (v3) | 1.0 |
|
| 37 |
+
| Checkpoint every | 500 epochs |
|
| 38 |
+
|
| 39 |
+
## Files
|
| 40 |
+
|
| 41 |
+
- `checkpoint-epoch-N.pt` — periodic checkpoints (model, optimizer, scheduler state)
|
| 42 |
+
- `final.pt` — final checkpoint after training
|
| 43 |
+
|
| 44 |
+
## Usage
|
| 45 |
+
|
| 46 |
+
```python
|
| 47 |
+
from huggingface_hub import hf_hub_download
|
| 48 |
+
import torch
|
| 49 |
+
from src.genome_minimizer_2.training.model import VAE
|
| 50 |
+
|
| 51 |
+
path = hf_hub_download("McClain/genome-minimizer-2", "final.pt", revision="v3")
|
| 52 |
+
checkpoint = torch.load(path, map_location="cpu")
|
| 53 |
+
|
| 54 |
+
model = VAE(input_dim=55039, hidden_dim=512, latent_dim=32)
|
| 55 |
+
model.load_state_dict(checkpoint["model_state_dict"])
|
| 56 |
+
```
|
| 57 |
+
|
| 58 |
+
## Links
|
| 59 |
+
|
| 60 |
+
- [W&B Experiment Tracking](https://wandb.ai/mcclain/genome-minimizer-2)
|
| 61 |
+
- [GitHub Repository](https://github.com/ucl-cssb/genome-minimizer-2)
|