Downloads · 30 days
0
vrptw-ddqn-alns
oggishi/vrptw-ddqn-alns
vrptw-ddqn-alns is a reinforcement learning model from oggishi. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. It is set up for safetensors. The card lists the license as mit.
Trained controller weights for the hybrid DDQN-ALNS solver in VRPTW-Research-Optimization. The network selects destroy/repair operators and acceptance behaviour inside an Adaptive Large Neighbourhood Search over the V…
Access
Public
Updated Jul 28, 2026
Repo size
753 KB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt402 KB · 53%
At a glance
- Task
- Reinforcement Learning
- Library
- safetensors
- License
- mit
- Access
- Public
- Created
- Jul 28, 2026
- Updated
- Jul 28, 2026
- SHA
- edbf2ac5
Model card
From the Hugging Face model README
DDQN-ALNS weights for VRPTW
Trained controller weights for the hybrid DDQN-ALNS solver in VRPTW-Research-Optimization. The network selects destroy/repair operators and acceptance behaviour inside an Adaptive Large Neighbourhood Search over the Vehicle Routing Problem with Time Windows.
Files
| File | Role |
|---|---|
rl_alns_dr_v15.safetensors | Domain-randomization checkpoint. Loaded by default. |
rl_alns_transfer.safetensors | Transfer-learning checkpoint (Solomon RC classes). |
gnn_edge_predictor.pt | Edge-scoring GNN used for repair heatmap guidance. |
Results
Under independent cold starts, the hybrid reaches the same minimum vehicle-count
floor as the ALNS baseline but hits it in 100% of runs versus 30-70% for the
baseline, and cuts total distance by 1.75%-4.07% at matched vehicle count on
Solomon and 200-customer Homberger instances. At 400 customers the vehicle-count
edge is small and only significant on two of three instances tested
(c2_4_1 p=0.0078, r2_4_1 p=0.0156; rc2_4_1 p=0.3750).
Usage
These are controller weights for a specific solver, not a standalone model. The production service downloads them at boot:
from huggingface_hub import hf_hub_download
path = hf_hub_download("oggishi/vrptw-ddqn-alns", "rl_alns_dr_v15.safetensors")
See solver_service/app.py in the repository for the full loading path.
- Task
- Reinforcement Learning
- Library
- safetensors
- License
- mit
- Created
- Jul 28, 2026
- Updated
- Jul 28, 2026