Data-driven Limited Area Mesoscale Prediction for Taiwan (DLAMP.tw)

https://github.com/yaochudosomething/dlamp
Admin
DLAMP.tw is a data-driven regional weather forecasting model for Taiwan that combines a...
Tokens:1,927
Snippets:16
Trust Score:5.1
License:Apache-2.0
Update:7 months ago
Show doc for...
Tokens:
Raw
### Setup Environment with UV

Source: https://github.com/yaochudosomething/dlamp/blob/master/README.md

Commands to install the UV package manager, clone the repository, create a virtual environment, and synchronize project dependencies.

```bash
curl -LsSf https://astral.sh/uv/install.sh | sh
echo 'eval "$(uv generate-shell-completion bash)"' >> ~/.bashrc
echo 'eval "$(uvx --generate-shell-completion bash)"' >> ~/.bashrc
source ~/.bashrc
git clone https://github.com/YaoChuDoSomething/dlamp.git
cd dlamp/
uv venv
source .venv/bin/activate
uv sync
```

--------------------------------

### Setup Environment with Conda and Pip

Source: https://github.com/yaochudosomething/dlamp/blob/master/README.md

Alternative installation method using Conda for environment management and Pip for installing NVIDIA Modulus and project-specific requirements.

```bash
conda create --name [env name] python=3.12 -y
conda activate [env name]
git clone https://github.com/NVIDIA/modulus.git && cd modulus
make install
pip3 install -r requirements.txt
pip3 install hydra-core --upgrade
pip install onnxruntime-gpu==1.23.2
```

--------------------------------

### Generate Project Context Commands

Source: https://github.com/yaochudosomething/dlamp/blob/master/doc/agentic-artifact/context_strategy.md

Commands to generate a codebase snapshot for AI analysis using either uv or pip, and a bash function for automated clipboard copying.

```bash
# Using uv
uvx gitingest . -o codebase.txt

# Using pip
gitingest . -o codebase.txt

# Bash alias for automated context generation and copying
function ai_context() {
    echo "Generating context for $(basename "$PWD")..."
    uvx gitingest . --exclude-pattern "*.lock,assets/*" -o /tmp/context.txt
    
    if command -v xclip &> /dev/null; then
        cat /tmp/context.txt | xclip -selection clipboard
        echo "Context copied to clipboard!"
    elif command -v pbcopy &> /dev/null; then
        cat /tmp/context.txt | pbcopy
        echo "Context copied to clipboard!"
    else
        echo "Context saved to /tmp/context.txt"
    fi
}
```

--------------------------------

### Execute Model Training and Inference

Source: https://github.com/yaochudosomething/dlamp/blob/master/README.md

Standard entry points for running the training pipeline and generating predictions using the configured model.

```bash
python train.py
python predict.py
```

--------------------------------

### Analyze heatwave data using GFS and ERA5 providers

Source: https://github.com/yaochudosomething/dlamp/blob/master/doc/refactor/oop.md

Demonstrates the initialization of weather data providers and the execution of heatwave analysis for specific geographic coordinates and dates. The implementation showcases dependency injection, allowing the WeatherAnalyzer to remain agnostic of the underlying data source.

```python
if __name__ == "__main__":
    # Define a date
    target_date = np.datetime64('2025-06-26')
    locations = [(25.03, 121.56), (35.68, 139.69)] # Taipei, Tokyo

    # 1. Use GFS Source
    # We can switch providers easily without changing the Analyzer logic (OCP).
    gfs = GFSProvider()
    analyzer_gfs = WeatherAnalyzer(gfs)
    print("GFS Results:", analyzer_gfs.analyze_heatwave(locations, target_date))

    # 2. Use ERA5 Source
    era5 = ERA5Provider()
    analyzer_era5 = WeatherAnalyzer(era5)
    print("ERA5 Results:", analyzer_era5.analyze_heatwave(locations, target_date))
```

--------------------------------

### Define AI System Prompt

Source: https://github.com/yaochudosomething/dlamp/blob/master/doc/agentic-artifact/context_strategy.md

A structured prompt template designed to frame the AI's persona and objectives when analyzing the generated codebase context.

```markdown
# Role
You are a Staff Software Engineer and Python Architect specializing in AI/ML pipelines.

# Objective
Analyze the attached codebase state for the project `dlamp-tw`.

# Guidelines
1. Token Efficiency: Do not repeat large blocks of code.
2. Context Awareness: Assume standard Python 3.10+.
3. Focus: Prioritize `src/` logic and `config/` structures.

# Current Task
[Insert your specific question here]

---
[PASTE GITINGEST OUTPUT HERE]
```

--------------------------------

### Configure Gitingest Ignore Patterns

Source: https://github.com/yaochudosomething/dlamp/blob/master/doc/agentic-artifact/context_strategy.md

Defines file patterns to exclude from AI context generation to save tokens and focus on relevant source code.

```text
# System & Environment
.git
.venv
__pycache__
*.pyc
.DS_Store
.env
.ipynb_checkpoints

# Large Info / Logs
logs/
outputs/
export/
gallery/
.gemini/

# Assets
assets/
*.png
*.jpg
*.jpeg
*.svg
*.gif
*.mp4
*.weights
*.pt
*.pth
*.onnx

# Lock files
uv.lock
package-lock.json
yarn.lock
```
Tokens:
Raw