The articles we publish cover a vertigo-inducing range of data science topics, but what we find even more impressive is just how much variety we find when we zoom in on a particular area. This week’s edition of the Variable, where we focus on the latest crop of machine learning must-reads, is a case in point. A good place to start is Josh Poduska’s comprehensive framework for maintaining ML models. Monitoring your model’s performance is one thing, Josh argues; approaching it holistically to ensure it performs well over time is another, and calls for constant iteration and creativity.
我们发表的文章涵盖了一系列令人眼花缭乱的数据科学主题，但我们发现更令人印象深刻的是，当我们放大某个特定领域时，我们发现的多样性是多么丰富。本周的《变量》(the Variable)就是一个很好的例子，我们将重点关注最新一批机器学习必读书籍。Josh Poduska的维护ML模型的综合框架是一个很好的开始。Josh认为，监控模型的性能是一回事;从整体上确保游戏在一段时间内运行良好是另一回事，这需要不断的迭代和创造力。
Photo by Taylor Heery on Unsplash
The most powerful posts on machine learning pull off something extremely difficult: translating complex ideas from the field’s cutting edge into narratives that are both accessible and energizing. Patrick Huembeli and his coauthors do just that when they explore the connections between energy-based models and the physics of interacting particles in order to better understand the inner workings of the former (the article is dedicated to Peter Wittek, their late collaborator). Felizia Quetscher sets a similar goal—transforming the dense and the opaque into something engaging and approachable—when she patiently explains convolutional neural networks (CNNs), and uses crisp illustrations to highlight the relationship of input and output arrays to the convolutional kernel between them. Finally, taking a cue from bioinformatics and computational biology, Remy Lau walks us through the theory and intuition behind network propagation, and uses the example of the HotNet2 algorithm to help make things more concrete.
关于机器学习的最具影响力的帖子完成了一件极其困难的事情:将该领域最前沿的复杂想法转化为既通俗易懂又充满活力的叙述。Patrick Huembeli和他的合作者正是这样做的，他们探索基于能量的模型和相互作用粒子的物理学之间的联系，以便更好地理解前者的内部工作原理(这篇文章是献给他们已故的合作者Peter Wittek的)。费利齐亚·奎彻(Felizia Quetscher)在耐心地解释卷积神经网络(CNN)时，也设定了类似的目标——将密集和不透明的东西转化为引人入胜和平易近人的东西，并用清晰的插图突出了输入和输出数组与它们之间的卷积核之间的关系。最后，以生物信息学和计算生物学为线索，Remy Lau向我们介绍了网络传播背后的理论和直觉，并使用HotNet2算法的例子来帮助使事情更加具体。
If you’re curious about some of the more practical, real-world stakes of machine learning, you’ll appreciate Mingjie Zhao’s ideas on structured thinking and how it can produce effective data storytelling. They cover all the extensive work data scientists need to perform to discover the modeling possibilities of the dataset they created. Along parallel lines, Javier Marin takes reinforcement learning from the realm of academic research into the startup scene, highlighting the potential of the armed-bandit algorithm to facilitate better business decision-making. Meanwhile, for Russo Alessio, machine learning is at once an academic field and a real-world issue: as an emerging scholar, he’s concerned about the effects of rapid growth and the field’s popularity on the quality and fairness of the conference peer-review process.
如果你对机器学习的一些更实际的、现实世界的利害关系感到好奇，你会欣赏Mingjie Zhao关于结构化思维的想法，以及它如何产生有效的数据故事。它们涵盖了数据科学家需要执行的所有广泛工作，以发现他们创建的数据集的建模可能性。与此类似，哈维尔·马林(Javier Marin)将强化学习从学术研究领域引入了创业领域，强调了武装强盗算法在促进更好的商业决策方面的潜力。与此同时，对于Russo Alessio来说，机器学习既是一个学术领域，也是一个现实世界的问题:作为一名新兴学者，他担心快速增长和该领域的普及对会议同行评审过程的质量和公平性的影响。
Had enough of machine learning for one week? Really? Well, that’s ok—we’ve got you covered, too:
一周的机器学习够了吗?真的吗?好吧，没关系，我们也会照顾你的。
Read our conversation with H20.ai’s Parul Pandey, where she talks about her transition from electrical engineering to data science, and how she broke down silos and found community along the way.
阅读我们与H20的对话。在那里，她谈到了自己从电气工程转向数据科学的过程，以及她如何打破孤岛，在这个过程中找到了社区。
Moran Beladev explains how to use NLP methods to represent a sequence of temporal graphs, whose topology evolves over time, with nodes and edges added and removed between different time snapshots.
Moran Beladev解释了如何使用NLP方法来表示时序图序列，其拓扑结构随着时间的推移而变化，在不同的时间快照之间添加和删除节点和边缘。
Finally, TDS Podcast host Jeremie Harris took part in a special crossover episode with the Banana Data podcast crew, discussing the issues surrounding AI and accountability.
最后，TDS播客主持人Jeremie Harris与Banana Data播客成员一起参加了一个特别的跨界节目，讨论了有关AI和问责制的问题。
We wish you all happy reading, and a good rest of the week—thank you for your support of our authors’ work.
我们祝大家阅读愉快，本周休息愉快——感谢您对我们作者工作的支持。
Until the next Variable,
直到下一个变量，
TDS Editors
原文出处：What’s Your Favorite Machine Learning Flavor? | by TDS Editors | Towards Data Science
