未加星标

I Deliver a Machine Learning Workshop to 120 People

字体大小 | |
[开发(python) 所属分类 开发(python) | 发布者 店小二05 | 时间 2019 | 作者 红领巾 ] 0人收藏点击收藏

I work in the Research division at a large tech company. One of the other divisions in my company asked me if I’d do a two-hour hands-on machine learning workshop. I would normally have to decline such a request because I’m super busy with my regular work duties. But in this case, I said yes because I have a similar workshop coming up and so I figured I’d use the requested workshop as a trial run for the other workshop.

Based on previous workshop experiences, I knew I wanted the attendees to install all the needed ML software during the workshop. Installing compatible versions of python, TensorFlow, Keras, PyTorch, CNTK, NumPy, and the dozens of other necessary systems is an absolute nightmare. Also, the installation process itself has all kinds of valuable information associated with it.

But installing all this software is difficult because almost everything assumes you have a live Internet connection to pull files as needed. My work building has extraordinarily good wireless network connectivity but even so, the installation file sizes can be extremely large and anything more than about a dozen people installing at once will bring the network to its knees.


I Deliver a Machine Learning Workshop to 120 People

I explained what neural networks are during the 15 minutes it took to install Anaconda.

So, a couple of weeks before the workshop, I set out to create an installation process that doesn’t require an Internet connection. That was quite a chore but I finally figured it all out.

So, the day of the workshop came and the room had over 120 people. I was a bit nervous but these guys were extremely sharp and they were able to follow along with the installation directions.

In the second hour, I did the classic Iris Dataset example using Keras. Some of my experienced colleagues scoff when I tell them I always start with Iris. They say, “Sheesh, everyone uses Iris.” And I reply that, yes, that’s exactly the point. The Iris data is used for a good reason ― it’s not too large and it provides a common example.


I Deliver a Machine Learning Workshop to 120 People

I don’t do it consciously, but I tend to wave my hands around a lot when I give a talk.

Anyway, the workshop went surprisingly well. My impression was the same as it always is when I do a beginner’s workshop ― the key is not what to teach, but what to leave out. Even the Iris program, which is only about 100 lines of code, is incredibly dense conceptually. For example, initializing the weights is one or two lines of code. But a full explanation of weight and bias initialization, even the five basic algorithms, (uniform, Gaussian, Glorot uniform, Glorot normal, He) could easily be a two-hour lecture all by itself. But beginners will get bogged to a halt if an instructor goes into too much detail.

Anyway, there’s no real moral to the story. Giving a training workshop or a lecture is formal education. But in some ways, most interactions we have with other people is informal education.


I Deliver a Machine Learning Workshop to 120 People
I Deliver a Machine Learning Workshop to 120 People

本文开发(python)相关术语:python基础教程 python多线程 web开发工程师 软件开发工程师 软件开发流程

代码区博客精选文章
分页:12
转载请注明
本文标题:I Deliver a Machine Learning Workshop to 120 People
本站链接:https://www.codesec.net/view/628467.html


1.凡CodeSecTeam转载的文章,均出自其它媒体或其他官网介绍,目的在于传递更多的信息,并不代表本站赞同其观点和其真实性负责;
2.转载的文章仅代表原创作者观点,与本站无关。其原创性以及文中陈述文字和内容未经本站证实,本站对该文以及其中全部或者部分内容、文字的真实性、完整性、及时性,不作出任何保证或承若;
3.如本站转载稿涉及版权等问题,请作者及时联系本站,我们会及时处理。
登录后可拥有收藏文章、关注作者等权限...
技术大类 技术大类 | 开发(python) | 评论(0) | 阅读(102)