Peter Song

Peter Song is a machine learning, data engineer, and AI writer with a focus on practical ML workflows. He shares insights on machine learning, data tools, and MLOps.

Using Transformers for Tabular Data Classification

by

When most people think of transformers in machine learning, they immediately picture natural language processing applications like ChatGPT or computer vision tasks with Vision Transformers. However, one of the most exciting and underexplored applications of transformer architecture lies in tabular data classification—a domain traditionally dominated by tree-based models like Random Forests and Gradient Boosting machines. … Read more

Multi-label Classification with scikit-learn

by

Multi-label classification represents one of the most challenging and practical problems in machine learning today. Unlike traditional single-label classification where each instance belongs to exactly one category, multi-label classification allows instances to be associated with multiple labels simultaneously. This approach mirrors real-world scenarios where data points naturally exhibit characteristics of multiple categories. Consider a movie … Read more

The Role of Feature Engineering in Deep Learning

by

In the rapidly evolving landscape of artificial intelligence, deep learning has emerged as a transformative force, powering everything from image recognition systems to natural language processing applications. However, beneath the sophisticated neural network architectures lies a fundamental question that continues to spark debate among data scientists and machine learning practitioners: What is the role of … Read more

How to Handle Missing Values in Time Series Forecasting

by

Missing values are one of the most common challenges data scientists face when working with time series data. Whether you’re analyzing stock prices, weather patterns, sensor readings, or sales figures, gaps in your data can significantly impact the accuracy and reliability of your forecasting models. Understanding how to properly identify, analyze, and handle these missing … Read more

How to Deploy LLMs on AWS Inferentia or GPU Clusters

by

Large Language Models (LLMs) have transformed the artificial intelligence landscape, but deploying these massive models efficiently in production remains one of the most significant technical challenges facing organizations today. With models like GPT-3, Claude, and Llama requiring substantial computational resources, choosing the right deployment infrastructure can make the difference between a cost-effective, scalable solution and … Read more

Using TensorFlow Data Pipelines for Large Datasets

by

When working with machine learning projects at scale, data preprocessing and loading often become the bottleneck that prevents models from reaching their full potential. TensorFlow’s tf.data API provides a powerful solution for building efficient data pipelines that can handle massive datasets while maintaining optimal performance. This comprehensive guide explores how to leverage TensorFlow data pipelines … Read more

Retraining Strategies for Online Machine Learning Systems

by

In today’s rapidly evolving digital landscape, machine learning systems must adapt continuously to changing data patterns, user behaviors, and business requirements. Unlike traditional batch learning approaches that retrain models on fixed datasets at predetermined intervals, online machine learning systems demand sophisticated retraining strategies that can handle streaming data while maintaining performance and stability. This article … Read more

Canary Deployments for Machine Learning Models

by

In the rapidly evolving landscape of machine learning operations (MLOps), deploying new models safely and efficiently has become a critical challenge that can make or break production systems. Traditional deployment strategies often involve significant risks, potentially exposing entire user bases to untested model behavior that could result in degraded performance, incorrect predictions, or complete system … Read more

Advantages of Transformer over LSTM in NLP Tasks

by

The field of Natural Language Processing (NLP) has witnessed a paradigm shift with the introduction of Transformer architecture in 2017. While Long Short-Term Memory (LSTM) networks dominated sequence modeling tasks for over two decades, Transformers have emerged as the superior choice for most NLP applications. Understanding the advantages of Transformer over LSTM in NLP tasks … Read more

Visualize Word2Vec Embeddings with t-SNE

by

Word embeddings have revolutionized how we represent language in machine learning, and Word2Vec stands as one of the most influential techniques in this space. However, understanding these high-dimensional representations can be challenging without proper visualization tools. This is where t-SNE (t-Distributed Stochastic Neighbor Embedding) becomes invaluable, offering a powerful way to visualize word2vec embeddings in … Read more