liuq4360 / xfer Goto Github PK

Website | Documentation | Contribution Guide

Xfer is a library that allows quick and easy transfer of knowledge^1,2,3 stored in deep neural networks implemented in MXNet. Xfer can be used with data of arbitrary numeric format, and can be applied to the common cases of image or text data.

Xfer can be used as a pipeline that spans from extracting features to training a repurposer. The repurposer is then an object that carries out predictions in the target task.

You can also use individual components of Xfer as part of your own pipeline. For example, you can leverage the feature extractor to extract features from deep neural networks or ModelHandler, which allows for quick building of neural networks, even if you are not an MXNet expert.

• Resource efficiency: you don't have to train big neural networks from scratch.
• Data efficiency: by transferring knowledge, you can classify complex data even if you have very few labels.
• Easy access to neural networks: you don't need to be an ML ninja in order to leverage the power of neural networks. With Xfer you can easily re-use them or even modify existing architectures and create your own solution.
• Utilities for feature extraction from neural networks.
• Rapid prototyping: ModelHandler allows you to easily modify a neural network architecture, e.g. by providing one-liners for adding / removing / freezing layers.
• Uncertainty modeling: With the Bayesian neural network (BNN) or the Gaussian process (GP) repurposers, you can obtain uncertainty in the predictions of the rerpurposer.

After defining an MXNet source model and data iterators for your target task, you can perform transfer learning with just 3 lines of code:

repurposer = xfer.LrRepurposer(source_model, feature_layer_names=['fc7'])
repurposer.repurpose(train_iterator)
predictions = repurposer.predict_label(test_iterator)

• Documentation
• Tutorial: Introduction and transfer learning for image data
• Tutorial: Transfer learning with automatic hyperparameter tuning
• Tutorial: Transfer learning for text data
• Tutorial: Creating your own custom repurposer
• Tutorial: xfer.ModelHandler for easy manipulation and inspection of MXNet models

• Dependencies: Primary dependencies are MXNet >=1.2 and GPy >= 1.9.5. See all requirements in setup.py.

• Supported architectures / versions: Python 3.6+ on MacOS and Amazon Linux.

  Also tested on Python 3.4 and 3.5. Kindly note that for these Python versions, NumPy must be pre-installed for GPy (dependency of Xfer) to work.

• Install with pip:

pip install xfer-ml

• Install from source: To install Xfer from source, after cloning the repository run the following from the top-level directory:

pip install .

To confirm installation, run:

>>> import xfer
>>> xfer.__version__

And confirm that version returned matches the expected package version number.

Have a look at our contributing guide, thanks for the interest!

Points of contact for Xfer are: Jordan Massiah, Keerthana Elango, Pablo G. Moreno, Nikos Aletras, Andreas Damianou

Xfer is licensed under the Apache 2.0 License.