Universal Timber Slab: Disciplinary Surrogate Models

<p>This dataset contains <strong>9 trained surrogate models</strong> across all four disciplines predicting the performance of UTS bay elements and a demo Python script.</p>

<h2>Model Artifacts</h2>

<p>Each surrogate is saved as a <code>.joblib</code> file which stores:</p>

<pre><code class="language-python">{
'model': <trained sklearn model>, # Trained model
'scaler': <StandardScaler or None>, # Feature scaler
'feature_names': List[str], # 31 feature names
'model_name': str, # e.g., 'Extra Trees'
'target': str, # Target variable
'discipline': str, # Discipline
'metrics': {
'test_r2': float,
'test_rmse': float,
'test_mae': float,
}
}
</code></pre>

<h2>Demo Python script</h2>

<p>The Python script <code>predict_bays.py</code> demonstrates how to extract the relevant features from the UTS BHoM data schema and use the trained models to predict the performance of each bay in a slab.</p>

<p>To run the CLI script, (<code>cd</code>) to the directory containing
<code>predict_bays.py</code>. By default, the trained models should be located in a <code>models</code> subdirectory within this directory. Alternatively, a custom
models directory can be specified using the <code>--models-dir</code> option.
</p>

<p>These standard scientific Python packages are required:</p>

<pre><code>pip install numpy pandas scikit-learn joblib</code></pre>

<p>The script can be run with multiple options:</p>

<pre><code>python predict_bays.py # default input
python predict_bays.py path/to/my_slab.json # custom input
python predict_bays.py input.json --format json # JSON output
python predict_bays.py input.json --bay 3 # single bay
python predict_bays.py input.json --live-load 3.0 # override defaults
python predict_bays.py input.json --models-dir ./models # custom models dir
</code></pre>