MetaNNvis
MetaNNvis is a tool for accessing introspection methods for neural networks regardless of the framework in which the neural network has been built. It is easily extendable and currently supports models from TensorFlow 2.0 and PyTorch in combination with methods from Captum and tf-keras-vis. For more details, see the project report.
Installation
The latest version of metaNNvis can be installed via pip:
pip install metaNNvis
Additionally, you need the following dependencies:
torch, seaborn, numpy, tensorflow, onnx2torch, onnx, captum, tf-keras-vis, torchvision, matplotlib
and onnx2keras. All dependencies can be installed via pip except for onnx2keras, which can be downloaded from the
project's GitHub page
Usage
For instructions on how to use cross-framework introspection and how to extend it by new methods, see the user guide.
Available methods
Cross-Framework Introspection currently supports most methods from Captum and all methods from tf-keras-vis. The supported methods are:
| Method | Category |
|---|---|
| Captum | |
| Integrated Gradients | primary, layer, neuron |
| Saliency | primary |
| DeepLift | primary, layer, neuron |
| GradientShap | primary, layer, neuron |
| Input X Gradient | primary |
| Gradient X Activation | layer |
| Deconvolution | primary, neuron |
| Feature Ablation | primary, layer, neuron |
| Feature Permutation | primary |
| Conductance | layer, neuron |
| Layer Activation | layer |
| GradCAM | layer |
| Neuron Gradient | neuron |
| tf-keras-vis | |
| Activation Maximization | feature visualization |
| Vanilla Saliency / SmoothGrad | attribution |
| GradCAM | attribution |
| GradCAM++ | attribution |
| ScoreCAM | attribution |
| LayerCAM | attribution |
Currently not supported are:
| Method | Category |
|---|---|
| Captum | |
| DeepLiftShap | primary, layer, neuron |
| Guided Backpropagation | primary, neuron |
| Guided GradCAM | primary |
| Occlusion | primary |
| Shapley Value Sampling | primary |
| Lime | primary |
| KernelShap | primary |
| Layer Relevance Propagation | primary, layer |
| Internal Influence | layer |
1 maintainer
Metadata
1 maintainer
Release files for metaNNvis 0.0.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| metannvis-0.0.3.tar.gz | 1.3 MB | Without provenance | Details for metannvis-0.0.3.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| metannvis-0.0.3-py3-none-any.whl | Python 3 | none | any | Details for metannvis-0.0.3-py3-none-any.whl |
Total release size: 1.3 MB
Release files / metannvis-0.0.3.tar.gz
| Download URL | metannvis-0.0.3.tar.gz |
|---|---|
| Size | 1.3 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
71e39dcf6872a907fa4398b55cb90587a001a92288f79eabaad99ef7db9730e6
|
|
BLAKE2b-256 checksum How to use checksums |
aaea2e0ea4bbc37950a4d8779f043ad237d4b620e3b5622a01539df17cb0f21c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.1 CPython/3.8.10
|
Release files / metannvis-0.0.3-py3-none-any.whl
| Download URL | metannvis-0.0.3-py3-none-any.whl |
|---|---|
| Size | 31.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
b21621dde1abaa508084fa3b7e536a17a10eeec8fcd4f2e32a32a2ef666b83ca
|
|
BLAKE2b-256 checksum How to use checksums |
6b57003b1938eea160c6d708cbb22b07569c7157df7ceca55a184214d09056e8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.1 CPython/3.8.10
|
Security for metaNNvis 0.0.3
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
