Awesome-OL User Guide
Comprehensive instructions for setting up and running experiments with Awesome-OL
Quick Start: For first-time users, we recommend beginning with the
main.ipynb notebook and following the demo section below.
🧪 Demo
Get started quickly with our interactive demo notebook:
- Locate
main.ipynbin the project root directory - Select your preferred framework, dataset, classifier, and strategies
- Customize hyperparameters as needed
- Optionally enable visualization for performance comparison
- Results are automatically saved in the
Resultsfolder
📂 Datasets
All datasets are stored as .csv files in the datasets folder with the following structure:
- Attributes: Feature columns for model input
- Labels: Target values for classification
Supported dataset formats:
['Jiaolong', 'Waveform', 'SEA', 'Hyperplane' or './datasets/' + name + '.csv']
Note: You can use any custom dataset in CSV format by placing it in the datasets folder and specifying the filename.
📈 Visualization Tools
Comprehensive visualization capabilities are included in the visualization folder:
- Multi-model confusion matrix: Compare classifier performance visually
- Dynamic GIFs: Track Accuracy curves and Macro F1 scores over time
- Interactive plots: Viewable directly in
main.ipynb
Example visualization outputs:
# Results include:
# - Accuracy/F1 GIF animations
# - Confusion matrix images
# - CSV files with numerical results
⚙️ Framework Choices
Select from four distinct framework approaches:
-
OL (Online Learning):
- Processes data one sample at a time
- Ideal for streaming data applications
-
ONE-STEP:
- Uses classifiers from
OAL_classifierfolder - Combines training and online learning in one phase
- Uses classifiers from
-
TWO-STEP-CHUNK:
- Uses classifiers from
classifierfolder - Separate training and online learning phases
- Supports batch processing
- Uses classifiers from
-
TWO-STEP-INSTANCE:
- Uses classifiers from
classifierfolder - Two-phase approach (training then online learning)
- Processes only one sample at a time
- Uses classifiers from
Selection Tip: Enter the number corresponding to your framework choice (1-4) when prompted in the notebook.
🔧 Experiment Parameters
Configure these essential parameters for your experiments:
| Parameter | Type | Description | Example Value |
|---|---|---|---|
max_samples |
Integer | Maximum number of samples to process | 10000 |
n_round |
Integer | Number of experimental rounds | 5 |
n_pt |
Integer | Number of performance tracking points | 20 |
dataset_name |
String | Name of dataset (from available options) | 'Waveform' |
Dataset Options: You can choose from predefined datasets or specify your own CSV file:
['Jiaolong', 'Waveform', 'SEA', 'Hyperplane' or './datasets/' + name + '.csv']
🤖 Model Selection Guide
Available models vary by framework choice:
OL Framework Models
['ARF', 'LB', 'OB', 'OBADWIN', 'DWM', 'OOB', 'SRP', 'AdaC2',
'QRBLS', 'BLS', 'OSSBLS', 'ISSBLS', 'SOSELM', 'NB', 'DES',
'DES_5', 'ACDWM', 'OLI2DS']
ONE-STEP Framework Models
['ROALE_DI', 'OALE']
TWO-STEP-CHUNK Framework
Available models:
['ARF', 'LB', 'OB', 'OBADWIN', 'DWM', 'OOB', 'SRP', 'AdaC2',
'QRBLS', 'BLS', 'OSSBLS', 'ISSBLS', 'SOSELM', 'NB', 'DES',
'DES_5', 'ACDWM', 'OLI2DS']
Available strategies:
['DMI_DD']
TWO-STEP-INSTANCE Framework
Available models:
['ARF', 'LB', 'OB', 'OBADWIN', 'DWM', 'OOB', 'SRP', 'AdaC2',
'QRBLS', 'BLS', 'OSSBLS', 'ISSBLS', 'SOSELM', 'NB', 'DES',
'DES_5', 'ACDWM', 'OLI2DS']
Available strategies:
['DSA_AI', 'Supervised', 'MTSGQS', 'US_fix', 'US_var',
'CogDQS', 'RS', 'DMI_DD']
Result Output: All predictions are saved as CSV files in the
Result folder. Visualization results are saved as GIFs (Accuracy/F1 curves) and confusion matrix images.