{"id":4578,"date":"2022-11-10T17:50:07","date_gmt":"2022-11-11T01:50:07","guid":{"rendered":"https:\/\/live-cometml.pantheonsite.io\/?p=4578"},"modified":"2025-04-24T17:16:31","modified_gmt":"2025-04-24T17:16:31","slug":"new-integration-comet-catalyst","status":"publish","type":"post","link":"https:\/\/www.comet.com\/site\/blog\/new-integration-comet-catalyst\/","title":{"rendered":"New Integration: Comet + Catalyst"},"content":{"rendered":"\n
\"\"\/<\/figure>\n\n\n\n

<\/p>\n\n\n\n

\n

We\u2019re excited to announce another excellent and powerful integration with Comet \u2014 Catalyst<\/a>! This integration allows you to leverage Comet\u2019s logging capabilities while using the Catalyst framework to structure your Deep Learning Experiment runs<\/p>\n<\/div>\n\n\n\n

\n

About Catalyst<\/h1>\n

Catalyst<\/a> is a PyTorch framework for Deep Learning R&D. It focuses on reproducibility, rapid experimentation, and codebase reusability so you can create something new rather than write yet another train loop. The Catalyst library incorporates research best practices so users can focus on building models and not worry about writing boilerplate code.<\/p>\n

About the Integration<\/h1>\n

Catalyst now ships with a dedicated CometLogger. With a slight modification (3 lines) of your Catalyst training code, you can now log metrics, hyperparameters, source code and much more from your runs to the Comet UI.<\/p>\n

Once you\u2019ve set up your account and configured your Comet API Key<\/a> within your project, simply pass the CometLogger to your Catalyst Trainer and you\u2019re good to go! You\u2019ll then also be able to take advantage of Comet\u2019s rich visualization capabilities<\/a><\/p>\n

import comet_ml\n\nimport os\nimport torch\nfrom torch import nn, optim\nfrom torch.utils.data import DataLoader\n\nfrom catalyst import dl\nfrom catalyst.data import ToTensor\nfrom catalyst.contrib.datasets import MNIST\nfrom torch.utils.data import DataLoader\n\n\nmodel = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))\ncriterion = nn.CrossEntropyLoss()\n\nlogger = dl.CometLogger()\n\nhparams = {\"lr\": 1.0e-3, \"batch_size\": 32}\noptimizer = optim.Adam(model.parameters(), lr=hparams[\"lr\"])\nloaders = {\n    \"train\": DataLoader(\n        MNIST(os.getcwd(), train=True, download=True, transform=ToTensor()),\n        batch_size=hparams[\"batch_size\"],\n    ),\n    \"valid\": DataLoader(\n        MNIST(os.getcwd(), train=False, download=True, transform=ToTensor()),\n        batch_size=hparams[\"batch_size\"],\n    ),\n}\n\nrunner = dl.SupervisedRunner(\n    input_key=\"features\", output_key=\"logits\", target_key=\"targets\", loss_key=\"loss\"\n)\n# model training\nrunner.train(\n    model=model,\n    criterion=criterion,\n    optimizer=optimizer,\n    loaders=loaders,\n    hparams=hparams,\n    num_epochs=1,\n    callbacks=[\n        dl.AccuracyCallback(\n            input_key=\"logits\", target_key=\"targets\", topk_args=(1, 3, 5)\n        ),\n        dl.PrecisionRecallF1SupportCallback(\n            input_key=\"logits\", target_key=\"targets\", num_classes=10\n        ),\n    ],\n    logdir=\".\/logs\",\n    valid_loader=\"valid\",\n    valid_metric=\"loss\",\n    minimize_valid_metric=True,\n    verbose=True,\n    load_best_on_end=True,\n    loggers={\"comet\": logger},\n)<\/pre>\n
Getting Started<\/h1>\n
Getting started with this integration is really easy \u2014<\/p>\n
The following resources should help you start logging Catalyst runs to Comet in no time:<\/p>\n