Image by Author
 

5 years ago, data scientists and machine learning engineers used to store Machine Learning (ML) experiment data on spreadsheets, paper, or on markdown files. Those days have long gone. Nowadays, we have highly efficient, user-friendly experiment tracking platforms. 

Apart from lightweight experiment tracking, these platforms come with data and model versioning, interactive dashboards, hyperparameter optimization, model registry, ML pipelines, and even model serving. 

In this post, we will be looking at the top 7 ML experiment tracking tools that are user-friendly, come with a lightweight API, and have an interactive dashboard to view and manage the experiments. 

 

1. MLflow Tracking

 

MLflow Tracking is a part of the open-source library MLflow. The API is used for logging experiments, metrics, parameters, output files, and code versions. The MLflow tracking also comes with web-based for you to visualize the result and interact with parameters and metrics. 

 

Image by Author
 

You can log queries and experiments using Python, R, Java, and REST API. MLflow also provides integration for popular ML frameworks such as Scikit-learn, Keras, PyTorch, XGBoost, and Spark.

 

2. DVC

 

Data Version Control · DVC is an open-source Git-based tool for versioning the data and model, ML pipelines, and ML experiment tracking. With Studio · DVC, you can log your experiments on a web application that provides UI for live experiment tracking, visualization, and collaboration.

 

Image from DVC
 

DVC is the ultimate tool that automates your workflow, stores, and version your data and model, provides CI/CD for ML, and simplifies your ML model deployments. You can access and store experiments using Python API, CLI, VSCode extension, and Studio. 

 

3. ClearML

 

ClearML Experiment tracks and automates everything related to ML operations. You can use it to log and share experiments, version the artifacts, and create ML pipelines. 

 

Image from ClearML
 

You can visualize the results, compare, reproduce, and manage all kinds of experiments. ClearML Experiment integrates with popular ML libraries such as PyTorch, TensorFlow, and XGBoost. You can get a basic version for free that covers all of the core processes of MLOps.  

 

4. DAGsHub

 

DAGsHub Logger allows you to log metrics, hyper parameters, and output using Python API. The Platform also supports experiment logging via MLflow which is quite useful for tracking live model performance.  

 

Image by Author
 

DagsHub provides you with code, data, and model versioning, experiments tracking, ML pipeline visualization, model serving, model monitoring, and team collaboration. It is a complete tool for your data science and machine learning projects.   

 

5. TensorBoard

 

TensorBoard is the first experimental logger that I have used. It is simple and integrates seamlessly with the TensorFlow package. By adding a few lines of code, you can track and visualize the metrics such as accuracy, loss, and F1 score over time.
 

Image from Documentation | TensorBoard
 

You can visualize the model graph, Projecting embeddings, view images, text, and audio data, and manage your experiment via TensorBoard UI. It is an easy, fast, and powerful tool. The downside is that it only works with the TensorFlow framework. 

 

6. Comet ML

 

Comet ML Experiment Tracking is a free ML experiment-tracking tool for the community. You can manage your experiment with a simple Python, Java, and R API that works with all of the popular machine learning frameworks, such as Keras, LightGBM, Transformers, and Pytorch.    
 

Image from Comet ML
 

You can log your experiments and view, compare, and manage them on a web application. The web user interface contains projects, reports, code, artifacts, models, and team collaboration. Moreover, you can modify your visualization or even create your graph using Python visualization libraries. 

 

7. Weights & Biases

 

Weights & Biases is a community-centric platform for tracking experiments, artifacts, interactive data visualization, model optimization, model registry, and workflow automation. With just a few lines of code, you can start tracking, comparing, and visualizing the ML models. 
 

Image from Weights & Biases
 

Apart from metrics, parameters, and outputs, you can also monitor CPU and GPU utilization, debug performance in real-time, store and version datasets up to 100 GB, and share your reports. 

 

Conclusion

 

All of these platforms are great. They come with some advantages and disadvantages. You can use them to showcase your portfolio, collaborate on projects, track experiments, and streamline the process to reduce human interference. 

Let me make things simple for you. 

• MLflow: open-source, free to use, and comes with all of the essential MLOps features. 
• DVC: it is for the people who are already using DVC for data and model versioning and want to use the same tool for ML pipelines and experiment tracking. 
• CLearML: end-to-end scalable MLOps platform. 
• DAGsHub: great for team collaboration, ML projects, and experiment tracking. It is also an end-to-end ML platform. 
• TensorBoard: if you are a fan or long-time user of TensorFlow, then for simplicity, use it for tracking the experiments. 
• Comet ML: this is a simple and interactive way of tracking ML models. 
• Weights & Biases: community-centric, simple, and recommended for ML portfolio. 

I hope you like my work. Do let me know in the comments if you have any questions or want to give me suggestions regarding MLOps tools. I know this space is saturated with ML tracking tools, and every other ML platform is now offering a lightweight experiment logging feature.
 
 
Abid Ali Awan (@1abidaliawan) is a certified data scientist professional who loves building machine learning models. Currently, he is focusing on content creation and writing technical blogs on machine learning and data science technologies. Abid holds a Master's degree in Technology Management and a bachelor's degree in Telecommunication Engineering. His vision is to build an AI product using a graph neural network for students struggling with mental illness.