Ensembling ConvNets using Keras
- Ensure you have Python 3 installed.
- Optional, but recommended: create a new virtual environment using your favourite virtual environment tool. Examle tools: virtualenv, pipenv.
- Install other dependencies:
pip install -r requirements.txt
jupyter lab keras_ensembling.ipynb. Browser window with Keras Ensembling Jupyter Lab notebook should come up automatically. If it doesn't come up, navigate to the running Jupyter Lab instance in your browser using the URL displayed in your terminal (i.e.
The Jupyter Notebook is running at: http://localhost:8888/).
- Training models in this notebook without a good GPU can take quite a long time. If you don't want to train models yourself, you can use pretrained ones. In this case, simply don't execute the cells where training happens, e.g:
_, conv_pool_cnn_weight_file = compile_and_train(conv_pool_cnn_model, NUM_EPOCHS)
Pretrained model weights will be loaded automatically wherever needed.
Copyright (c) 2018 - 2019 Maxim Mikhaylov