odin-ai icon indicating copy to clipboard operation
odin-ai copied to clipboard

verified publisher icon trungnt13
→

Metadata

Orgainzed Digital Intelligent Network (O.D.I.N)

.. image:: https://readthedocs.org/projects/odin/badge/ :target: http://odin0.readthedocs.org/en/latest/

O.D.I.N

Organized Digital Intelligent Network (O.D.I.N)

O.D.I.N is a framework for building "Organized Digital Intelligent Networks".

End-to-end design, versatile, plug-n-play, minimized repetitive work

This repo contains the most comprehensive implementation of variational autoencoder and disentangled representation benchmark.

.. code-block:: python

from odin.fuel import MNIST from odin.networks import get_networks from odin.bay.vi import VariationalAutoencoder

ds = MNIST() train = ds.create_dataset(partition='train')

optimized architectures for MNIST

networks = get_networks(ds, is_hierarchical=False, is_semi_supervised=False)

create the VAE

vae = VariationalAutoencoder(**networks) vae.build(ds.full_shape) vae.fit(train, max_iter=10000)

TOC

  1. VAE__
  2. Hierachical VAE__
  3. Semi-supervised VAE__
  4. Disentanglement Gym__
  5. Faster Classical ML__ (automatically select GPU implementation)

.. __: #variational-autoencoder-vae .. __: #hierarchical-vae .. __: #semi-supervised-vae .. __: #disentanglement-gym .. __: #fast-api-for-classical-ml

Variational Autoencoder (VAE)

.. list-table:: :widths: 30 80 25 :header-rows: 1

    • Model
    • Reference/Description
    • Implementation
      1. Vanilla VAE
    • (Kingma et al. 2014). "Auto-Encoding Variational Bayes" [Paper <https://arxiv.org/abs/1312.6114>_]
    • [Code <https://github.com/trungnt13/odin-ai/blob/5c83586999a15a02ebbcb7b5f7336f1cce245ae4/odin/bay/vi/autoencoder/variational_autoencoder.py#L132>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/vae_basic_test.py>]
      1. Beta-VAE
    • (Higgins et al. 2016). "beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework" [Paper <https://openreview.net/forum?id=Sy2fzU9gl>_]
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/beta_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/unsupervised_vae_test.py>]
      1. BetaGamma-VAE
    • Customized version of Beta-VAE, support re-weighing both reconstruction and regularization \(\mathrm{ELBO}=\gamma \cdot E_q[log p(x|z)] - \beta \cdot KL(q(z|x)||p(z|x))\)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/beta_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. Annealing VAE
    • (Sønderby et al. 2016) "Ladder variational autoencoder"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/beta_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. CyclicalAnnealing VAE
    • (Fu et al. 2019) "Cyclical Annealing Schedule: A Simple Approach to Mitigating KL Vanishing"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/beta_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. BetaTC-VAE
    • (Chen et al. 2019) "Isolating Sources of Disentanglement in Variational Autoencoders" (regularize the latents' Total Correlation)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/beta_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. Controlled Capacity Beta-VAE
    • (Burgess et al. 2018) "Understanding disentangling in beta-VAE"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/beta_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. FactorVAE
    • (Kim et al. 2018) "Disentangling by Factorising"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/factor_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. AuxiliaryVAE
    • (Maaløe et al. 2016) "Auxiliary Deep Generative Models"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/auxiliary_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. HypersphericalVAE
    • (Davidson et al. 2018) "Hyperspherical Variational Auto-Encoders"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/hyperbolic_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. PowersphericalVAE
    • (De Cao et al. 2020) "The Power Spherical distribution"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/hyperbolic_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. DIPVAE
    • (Kumar et al. 2018) "Variational Inference of Disentangled Latent Concepts from Unlabeled Observations" (I - only_mean=True; II - only_mean=False)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/dip_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. InfoVAE
    • (Zhao et al. 2018) "infoVAE: Balancing Learning and Inference in Variational Autoencoders"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/info_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. MIVAE
    • (Ducau et al. 2017) "Mutual Information in Variational Autoencoders" (max Mutual Information I(X;Z))
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/info_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. irmVAE
    • (Jing et al. 2020) "Implicit Rank-Minimizing Autoencoder" (Implicit Rank Minimizer)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/irm_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. ALDA
    • (Figurnov et al. 2018) "Implicit Reparameterization Gradients" (Amortized Latent Dirichlet Allocation - VAE with Dirichlet latents for topic modeling)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/lda_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. TwoStageVAE
    • (Dai et al. 2019) "Diagnosing and Enhancing VAE Models"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/two_stage_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. VampriorVAE
    • (Tomczak et al. 2018) "VAE with a VampPrior"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/vamprior.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]
      1. VQVAE
    • (Oord et al. 2017) "Neural Discrete Representation Learning"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/vq_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/betavae_encoder_info_bound.py>]

Hierarchical VAE

.. list-table:: :widths: 30 80 25 :header-rows: 1

    • Model
    • Reference/Description
    • Implementation
      1. LadderVAE
    • (Sønderby et al. 2016) "Ladder variational autoencoder"
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/hierarchical_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/vae_basic_test.py>]
      1. BidirectionalVAE
    • (Kingma et al. 2016) "Improved variational inference with inverse autoregressive flow" (Bidirectional inference hierarchical VAE)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/hierarchical_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/vae_basic_test.py>]
      1. ParallelVAE
    • (Zhao et al. 2017) "Learning Hierarchical Features from Generative Models" (Multiple latents connects encoder-decoder from bottom to top in parallel)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/hierarchical_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/vae_basic_test.py>]

Semi-supervised VAE

.. list-table:: :widths: 30 80 25 :header-rows: 1

    • Model
    • Reference/Description
    • Implementation
      1. Semi-supervised FactorVAE
    • Same as FactorVAE, but the discriminator also estimate the density of the labels and unlabeled data (like in semi-GAN)
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/factor_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/semafo_final.py>]
      1. MultiheadVAE
    • VAE has multiple decoders for different tasks
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/multitask_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/semafo_final.py>]
      1. SkiptaskVAE
    • VAE has multiple tasks directly constrain the latents
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/multitask_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/semafo_final.py>]
      1. ConditionalM2VAE
    • (Kingma et al. 2014) "Semi-supervised learning with deep generative models" [Paper <https://arxiv.org/abs/1406.5298>_]
    • [Code <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/autoencoder/conditional_vae.py>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/semafo_final.py>]
      1. CCVAE (capture characteristic VAE)
    • (Joy et al. 2021) "Capturing label characteristics in VAEs" [Paper <https://openreview.net/forum?id=wQRlSUZ5V7B>_]
    • [Code <https://github.com/trungnt13/odin-ai/blob/aea88577cbc972230e3d9dabfbe6144509364768/examples/vae/semafo_final.py#L1130>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/semafo_final.py>]
      1. SemafoVAE
    • (Trung et al. 2021) "The transitive information theory and its application to deep generative models" [Paper <github.com/trungn13>_]
    • [Code <https://github.com/trungnt13/odin-ai/blob/aea88577cbc972230e3d9dabfbe6144509364768/examples/vae/semafo_final.py#L351>][Example <https://github.com/trungnt13/odin-ai/blob/master/examples/vae/semafo_final.py>]

Disentanglement Gym

DisentanglementGym <https://github.com/trungnt13/odin-ai/blob/master/odin/bay/vi/disentanglement_gym.py>_: fast API for benchmarks on popular datasets and renowned disentanglement metrics.

Dataset support: ['shapes3d', 'dsprites', 'celeba', 'fashionmnist', 'mnist', 'cifar10', 'cifar100', 'svhn', 'cortex', 'pbmc', 'halfmoons']

Metrics support:

  • Correlation: 'spearman', 'pearson', 'lasso'
  • BetaVAE score
  • FactorVAE score
  • Mutual Information Estimated
  • MIG (Mutual Information Gap)
  • SAP (Separated Attribute Prediction)
  • RDS (relative disentanglement strength)
  • DCI (Disentanglement, Completeness, Informativeness)
  • FID (Frechet Inception Distance)
  • Total Correlation
  • Clustering scores: Adjusted Rand Index, Adjusted Mutual Info, Normalized Mutual Info, Silhouette score.

Fast API for classical ML

Automatically accelerated by RAPIDS.ai (i.e. automatically select GPU implementation if available)

Dimension Reduction


* t-SNE [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/fast_tsne.py>`_]
* UMAP [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/fast_umap.py>`_]
* PCA, Probabilistic PCA, Supervised Probabilistic PCA, MiniBatch PCA, Randomize PCA [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/decompositions.py>`_]
* Probabilistic Linear Discriminant Analysis (PLDA) [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/plda.py>`_]
* iVector (GPU acclerated) [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/ivector.py>`_]

GMM
~~~

* GMM classifier [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/gmm_classifier.py>`_]
* Probabilistic embedding with GMM [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/gmm_embedding.py>`_]
* Universal Background Model (GMM-Tmatrix) [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/gmm_tmat.py>`_]

Clustering
~~~~~~~~~~

* KNN [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/cluster.py>`_]
* KMeans [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/cluster.py>`_]
* DBSCAN [`Code <https://github.com/trungnt13/odin-ai/blob/master/odin/ml/cluster.py>`_]

← Metadata

21
Stars
11
Forks
Watchers

Owner

verified publisher icon trungnt13

Metadata

Orgainzed Digital Intelligent Network (O.D.I.N)

Back