Copyright thenewstack
Python has long been the dominant language for training and using machine learning models, but that’s changing. “We’re bringing machine learning tools into the JavaScript ecosystem,” Laurie Lay, a senior software engineer with Ippon Technologies, said. “By doing that, we are making this technology accessible to the largest and most active developer community in the world, which is the JavaScript community.” At this year’s devmio International JavaScript Conference, Lay demoed five open source JavaScript libraries available now for frontend and web developers who want to dive into machine learning. Lay specializes in full-stack development and is a part of Ippon Technology’s AI Center of Excellence. These five libraries give JavaScript developers a way to start using machine learning and working with models in JavaScript, she said. 1. Danfo.js Python’s Pandas can clean, transform and structure data, Lay said. Pandas is basically a Python wrapper around C that makes data manipulation a lot easier. In the JavaScript ecosystem, there’s Danfo.js, which is “heavily inspired by Panda,” she said. She outlined what Danfo.js offers: Data manipulation and processing; Preparation and cleaning of data before training the model; Pandas-like API, easy data wrangling that integrates with TensorFlow.js. “It introduces the data frame and the series data structures to JavaScript, which are used for handling relational and labeled data,” she said. “Danfo.js is just for understanding what’s going on in your data, being able to understand [if] there are any deviations or any kind of outliers that you would need to go back in and fix.” It also has a VS code extension, she added. 2. The Natural Library The Natural Library is a lightweight tool for natural language processing, Lay said. She noted that The Natural Library offers: Natural language processing (NLP), including tokenization, splitting text into words or stemming, which is reducing words down to their root form; Simple API for tokenizing, stemming, classification and sentiment analysis; Quick and effective text-based ML tasks. 3. Synaptic Synaptic is a JavaScript library for building neural networks. “This is the one that I was saying that’s a little bit easier for building neural networks within JavaScript, because it doesn’t require you to have any of these other languages, like Python,” Lay said. “It’s really easy to set up a synaptic neural network.” Lay’s slides showed Synaptic is used for: Neural networks; Architecture-free, highly modular and supports complex network types; and Experimenting with custom neural network architectures. 4. TensorFlow.js TensorFlow.js is an open source JavaScript library that allows developers to use and build ML models directly in the web browser or within a Node.js environment. “If you need to perform complex tasks like image or audio classification, then you would want to leverage the powerful, pre-trained models of TensorFlow.js,” Lay said. “This is the undisputed heavyweight champion of the production-grade deep learning.” TensorFlow.js is used for: General purpose and deep learning; Powerful GPU-accelerated large ecosystems, pre-trained models; and Heavy-duty model training, deep learning, image/audio tasks. 5. Scikit.js Scikit.js is used for predictive data analysis and machine learning. It aims to be a TypeScript port of the scikit-learn python library, according to its npm notes. “Lastly, if you’re looking to leverage the classical machine learning algorithms that you might be familiar with from Python’s machine learning ecosystem, or if you’re comfortable with scikit-learn API in Python, then you might want to pick up something like Scikit.js, which is an almost identical API,” Lay said. Use Scikit.js for: Traditional ML models; Familiar scikit-learn API, broad range of classical algorithms; Developers transitioning from Python’s scikit-learn. “These tools are available now,” Lay said. “The community is expanding, and the best way to take advantage of that is to pick up one of the libraries, find a data set that interests you and start building because we as JavaScript developers can shape that intelligent, data-driven web future.” Postscript: There is a sixth option for working with machine learning models in the browser, as Lay’s Ippon Technologies colleague, Julian Wilkison-Duran, shared at the same conference. Read more about it in “JavaScript Library Runs Machine Learning Models in Browser.”
