In this paper, we propose a new approach from the sample-level classification difficulty identifying, sampling and ensemble learning. Accordingly, we design an ensemble approach in pipe with sample-level gradient resampling, i.e., Balanced Cascade with Filters (BCWF). Before that, as a preliminary exploration, we first design a Hard Examples Mining Algorithm (HEM) to explore the gradient distribution of classification difficulty of samples and identify the hard examples.
The figure below gives an overview of the our framework.
**Main dependencies: **
- Python (>=3.5)
- pandas (>=0.23.4)
- numpy (>=1.11)
- scikit-learn (>=0.20.1)
- imbalanced-learn (=0.5.0, optional, for baseline methods)
To install requirements, run:
pip install -r requirements.txtA typical usage example:
# Define model
model_class = BcwfH(dataset_name, T=15)
# Model calculation
model_class.apply_all()
# Metrics
metrics = model_class.display()You can run .py file too.
Here is an example:
# Run Model
python main.py
# Hyper-test
python hyper.py
# Get comparsion
python result.pyThanks goes to these wonderful people (emoji key):
 Chuang Zhao 🤔 💻 |
Nanlin Liu 🤔 💻 |
This project follows the all-contributors specification. Contributions of any kind welcome!