{"id":4946,"date":"2023-01-03T11:33:03","date_gmt":"2023-01-03T19:33:03","guid":{"rendered":"https:\/\/live-cometml.pantheonsite.io\/?p=4946"},"modified":"2025-04-24T17:16:15","modified_gmt":"2025-04-24T17:16:15","slug":"pythae-comet","status":"publish","type":"post","link":"https:\/\/www.comet.com\/site\/blog\/pythae-comet\/","title":{"rendered":"Pythae +\u00a0Comet"},"content":{"rendered":"\n
\n
\n
\n<\/div>\n
\n
\"Two<\/figure>
\n

The Pythae<\/strong> library<\/a>, which brings together many Variational Autoencoder models and enables researchers to make comparisons and conduct reproducible research, is now integrated with Comet ML<\/strong>!<\/p>\n

<\/figure>\n

The Comet ML<\/a> experiment tracking tool is very useful for researchers to store their experiment configs, track their training, and compare the results in an easy and understandable way through a visual interface.<\/p>\n

Now let\u2019s see in practice how to easily monitor an experiment with Comet ML in Pythae!<\/p>\n

What is a (Variational) Autoencoder?<\/h3>\n

Images, texts, sounds, and more, produced realistically with deep neural networks, have come to the fore in recent years as the output of surprisingly talented models.<\/p>\n

Although these generative models, which are well-designed and require huge data, appear in the literature with many different architectures, the Generative Attractive Networks and Variable Autoencoders model families are at the forefront of this race!<\/p>\n

Autoencoders are non-generative models that aim to automatically learn to convert any data to code, consisting of two basic parts, an Encoder and a Decoder. Its main purpose is to compress the data given as input and reproduce it with as little loss as possible. The Variational Autoencoder family, on the other hand, includes generator models that have the ability to generate random codes by sampling and obtain new data with this code.<\/p>\n

\n<\/figure><\/div><\/div><\/section>\n\n\n\n
\"Two
Denoising with Auto Encoders – Keras<\/figcaption><\/figure>\n\n\n\n

The idea of Variational Autoencoder was presented in 2013 by Diederik P. Kingma and Max Welling in the article \u201cAuto-Encoding Variational Bayes.\u201d<\/strong><\/a><\/p>\n\n\n\n

\n<\/figure>\n\n\n\n
\"A
A continuous space of faces generated by Tom White using VAEs. Source: https:\/\/gaussian37.github.io\/deep-learning-chollet-8-4\/<\/figcaption><\/figure>\n\n\n\n

What distinguishes the models in this family from standard Autoencoders is that the input passed through the Encoder is encoded as a probability distribution. If this probability distribution is, for example, a normal distribution, the Encoder output will be the mean and variance values. By sampling these values, the code is obtained and this code can be solved with the help of a Decoder. Although the Decoder structure is the same in Standard and Variational Autoencoder models, there is a difference in the Encoder structure.<\/p>\n\n\n\n

<\/figure>\n\n\n\n
\"an
Auto-Encoder vs Variational Auto-Encoder [Kaynak]<\/figcaption><\/figure>\n\n\n\n

With Auto Encoder, they can be trained easily with a single loss function without the need for extra parameters. In Variational Encoders, on the other hand, it is difficult and challenging to find the balance between reconstruction and latent loss, so training is quite difficult. However, the fact that Auto Encoders are prone to overfitting causes Variational Auto Encoders to be preferred.<\/p>\n\n\n\n

What is Pythae?<\/h3>\n\n\n\n

Pythae is a Python library that gathers the most commonly used (Variational) Auto-Encoder models under one roof, providing easy benchmarking experiments and comparison opportunities, especially for researchers.<\/p>\n\n\n\n

It allows researchers to train any of the models it contains with their own data and allows the sharing and uploading of existing models on HuggingFace Hub. On the other hand, thanks to its integration with tools such as Comet, it allows you to monitor the experiments.<\/p>\n\n\n\n

\ud83d\udfe0 You can install the latest stable version of the pythae library using pip:<\/strong> pip install pythae<\/code><\/p>\n\n\n\n

\ud83d\udfe0To install the latest version of the Pythae library: <\/strong>pip install git+https:\/\/github.com\/clementchadebec\/benchmark_VAE.git<\/code><\/p>\n\n\n\n

Some models in the Pythae library:<\/strong> Autoencoder (AE), Variational Autoencoder (VAE), Beta Variational Autoencoder (BetaVAE), VAE with Linear Normalizing Flows (VAE_LinNF), VAE with Inverse Autoregressive Flows (VAE_IAF), Disentangled Beta Variational Autoencoder (DisentangledBetaVAE), Disentangling by Factorising (FactorVAE), Beta-TC-VAE (BetaTCVAE).<\/p>\n\n\n\n

Some samplers available with the Pythae library: <\/strong>Normal prior (NormalSampler), Gaussian mixture (GaussianMixtureSampler), Two stage VAE sampler (TwoStageVAESampler), Unit sphere uniform sampler (HypersphereUniformSampler), Poincar\u00e9 Disk sampler (PoincareDiskSampler), VAMP prior sampler (VAMPSampler).<\/p>\n\n\n\n

You can find the complete list of models and samplers and all sample source codes for the library here.<\/strong><\/a><\/p>\n\n\n\n

What is Comet?<\/h3>\n\n\n\n

Comet is a platform that enables teams and individuals to manage key machine learning lifecycle steps such as monitoring, versioning, model registration, and comparing results, especially iterative model training processes performed by teams of data science and machine learning researchers. And now Comet is fully integrated with Pythae!<\/p>\n\n\n\n

\"an<\/figure>\n\n\n\n

 <\/p>\n\n\n\n

The Comet platform supports every stage of the machine learning lifecycle, from monitoring training runs to monitoring models in production. In addition, due to its flexible nature, it can be run with dozens of installation options both for the company and on any infrastructure, including virtual private cloud (VPC) installations. On the other hand, you can launch Comet easily and quickly by adding just two lines of code to your script, notebook, or pipeline. This will ensure that everything you need to monitor and manage your code, metrics and hyperparameters will be transferred to the platform.<\/p>\n\n\n\n

Comet is free for individuals and you can create your account here.<\/a><\/p>\n\n\n\n

The easiest way to install Comet on your system is to use <\/strong>pip<\/strong><\/code>: <\/strong>pip install comet_ml<\/code><\/p>\n\n\n\n

\n
\n
\n

Pythae + Comet<\/h3>\n

Now that we have an idea about Pythae and Comet, it\u2019s time to practice!<\/p>\n

The Variational Autoencoder (VAE) model has been implemented in the Pythae<\/strong><\/a> library. We will also train the reconstruction example in the MNIST dataset using the Pythae library and see how we can follow the training logs via Comet ML<\/strong><\/a>.<\/strong><\/p>\n

Importing Libraries<\/strong><\/h4>\n

In PyTorch, you can import the torchvision <\/em><\/strong>package to use the vision datasets, the pythae<\/em><\/strong> package to train the reconstruction model practically, the CometCallback function of Pythae to monitor the logs, which includes the integration of Pythae and CometML, and the comet_ml<\/em><\/strong> package if you want to show the Comet UI directly in the Jupyter notebook.<\/p>\n

import torchvision.datasets as datasets\n\n#pythae\nfrom pythae.models import BetaVAE, BetaVAEConfig\nfrom pythae.trainers import BaseTrainerConfig\nfrom pythae.pipelines.training import TrainingPipeline\nfrom pythae.models.nn.benchmarks.mnist import Encoder_ResNet_VAE_MNIST, Decoder_ResNet_AE_MNIST\n\n# Create you callback\nfrom pythae.trainers.training_callbacks import CometCallback\n\n# Or you can alternatively ability to view the Comet UI in the jupyter notebook\nimport comet_ml<\/span><\/pre>\n
Downloading the Dataset<\/h4>\n
Let\u2019s download the MNIST dataset with the torchvision.datasets<\/em><\/strong> package. Then, let\u2019s reshape it according to the input shape of the model by separating it as train and eval. Finally, let\u2019s normalize.<\/p>\n
mnist_trainset = datasets.MNIST(root='..\/data', train=True, download=True, transform=None)\n\ntrain_dataset = mnist_trainset.data[:-10000].reshape(-1, 1, 28, 28) \/ 255.\neval_dataset = mnist_trainset.data[-10000:].reshape(-1, 1, 28, 28) \/ 255.<\/span><\/pre>\n
Defining Model Parameters<\/h4>\n