A library for training multiclass classifiers with weak labels
python3.7 -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
See example of a call inside one of the add_queue scripts
Run the unit tests with the make command
make testIn order to run any notebook in this repository, first create a new kernel that will be available from the Jupyter Notebook.
# Load the virtual environment
source venv/bin/activate
# Create a new kernel
ipython kernel install --name "weaklabels" --user
# Open Jupyter
jupyter notebook
After opening or creating a notebook, you can select the "weaklabels" kernel in kernel -> Change kernel -> weaklabels
Current usage (may need updating)
Usage: main.py [options]
Options:
-h, --help show this help message and exit
-p DATASETS, --datasets=DATASETS
List of datasets or toy examples totest separated by
with no spaces.
-s NS, --n-samples=NS
Number of samples if toy dataset.
-f NF, --n-features=NF
Number of features if toy dataset.
-c N_CLASSES, --n-classes=N_CLASSES
Number of classes if toy dataset.
-m N_SIM, --n-simulations=N_SIM
Number of times to run every model.
-l LOSS, --loss=LOSS Loss function to minimize between square (brier score)
or CE (cross entropy)
-u PATH_RESULTS, --path-results=PATH_RESULTS
Path to save the results
-r RHO, --rho=RHO Learning step for the Gradient Descent
-a ALPHA, --alpha=ALPHA
Alpha probability parameter
-b BETA, --beta=BETA Beta probability parameter
-g GAMMA, --gamma=GAMMA
Gamma probability parameter
-i N_IT, --n-iterations=N_IT
Number of iterations of Gradient Descent.
-e MIXING_MATRIX, --mixing-matrix=MIXING_MATRIX
Method to generate the mixing matrix M.One of the
following: IPL, quasi-IPL, noisy, random_noise,
random_weak
The python code in utils/data.py can be run in order to check that all datasets can be downloaded, preprocessed and a classifier can be trained on them. This can be done by calling the python code as a main file
python utils/data.py
Should output the following for each dataset
Testing all datasets
Evaluating iris[61] dataset
----------------
Dataset description:
Dataset name: iris
Sample size: 150
Number of features: 4
Number of classes: 3
Logistic Regression score = 0.9733333333333334
Test and upload the code with the make command
make pypi
It may require user and password if these are not set in your home directory a file .pypirc
[pypi]
username = __token__
password = pypi-yourtoken
- Jesus Cid Sueiro
- Miquel Perello Nieto
- Daniel Bacaicoa