|
|
---
|
|
|
library_name: pytorch
|
|
|
tags:
|
|
|
- graph-neural-network
|
|
|
- knowledge-graph
|
|
|
- agricultural-ai
|
|
|
- crop-recommendation
|
|
|
- gcn
|
|
|
- graph-embeddings
|
|
|
license: mit
|
|
|
datasets:
|
|
|
- ugandan-agricultural-data
|
|
|
---
|
|
|
|
|
|
# Agricultural AI Graph Embedding Models
|
|
|
|
|
|
Graph neural network models trained on Ugandan agricultural knowledge graph for crop recommendation.
|
|
|
|
|
|
## Model Overview
|
|
|
|
|
|
This repository contains multiple graph embedding models trained on an agricultural knowledge graph with 175,318 triples representing crop-soil-climate relationships.
|
|
|
|
|
|
## Models Included
|
|
|
|
|
|
### Best Model: GCN (Graph Convolutional Network)
|
|
|
- **File**: `best_model.pth`
|
|
|
- **Accuracy**: 87.28%
|
|
|
- **F1-Score**: 85.71%
|
|
|
- **ROC-AUC**: 96.90%
|
|
|
- **Embedding Dimension**: 100
|
|
|
- **Entities**: 2,513
|
|
|
- **Relations**: 15
|
|
|
|
|
|
### Individual Models
|
|
|
1. **GCN Model** (`gcn_model.pth`) - Best performing
|
|
|
2. **TransE Model** (`transe_model.pth`) - Translation-based
|
|
|
3. **DistMult Model** (`distmult_model.pth`) - Bilinear
|
|
|
4. **ComplEx Model** (`complex_model.pth`) - Complex embeddings
|
|
|
5. **GraphSAGE Model** (`graphsage_model.pth`) - Sampling-based
|
|
|
|
|
|
## Model Metadata
|
|
|
|
|
|
The `model_metadata.json` file contains:
|
|
|
- Entity to ID mappings (2,513 entities)
|
|
|
- Relation to ID mappings (15 relations)
|
|
|
- ID to entity mappings
|
|
|
- Model configuration parameters
|
|
|
|
|
|
## Usage
|
|
|
|
|
|
```python
|
|
|
import torch
|
|
|
from huggingface_hub import hf_hub_download
|
|
|
|
|
|
# Download model
|
|
|
model_path = hf_hub_download(
|
|
|
repo_id="Awongo/soil-crop-recommendation-model",
|
|
|
filename="best_model.pth"
|
|
|
)
|
|
|
|
|
|
# Download metadata
|
|
|
metadata_path = hf_hub_download(
|
|
|
repo_id="Awongo/soil-crop-recommendation-model",
|
|
|
filename="model_metadata.json"
|
|
|
)
|
|
|
|
|
|
# Load model (pseudo-code - adjust to your model architecture)
|
|
|
# model = GCNModel(num_entities=2513, num_relations=15, embedding_dim=100)
|
|
|
# model.load_state_dict(torch.load(model_path, map_location='cpu'))
|
|
|
# model.eval()
|
|
|
```
|
|
|
|
|
|
## Training Data
|
|
|
|
|
|
- **Knowledge Graph**: 175,318 triples
|
|
|
- **Dataset**: Ugandan agricultural data
|
|
|
- **Literature**: 52 research papers
|
|
|
- **Crops**: 8 major crops (maize, rice, beans, cassava, sweet potato, banana, coffee, cotton)
|
|
|
|
|
|
## Application
|
|
|
|
|
|
Used in production for agricultural crop recommendations based on:
|
|
|
- Soil properties (pH, organic matter, nutrients)
|
|
|
- Climate conditions (temperature, rainfall)
|
|
|
- Knowledge graph embeddings
|
|
|
|
|
|
## Citation
|
|
|
|
|
|
```bibtex
|
|
|
@misc{agricultural-ai-graph-models,
|
|
|
title={Agricultural AI Graph Embedding Models for Crop Recommendation},
|
|
|
year={2025},
|
|
|
publisher={Hugging Face}
|
|
|
}
|
|
|
```
|
|
|
|
