<p><center><a href="https://www.datacamp.com/jobs/"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1602862294/James_spotlight_w2g0zg.png" /></a></center><br></p>
<h2 id="what-was-your-journey-to-joining-datacamp-">What was your journey to joining DataCamp?</h2>
<p>I was fortunate to have found DataCamp right when I finished my studies at Hobart and William Smith Colleges (HWS). I learned about DataCamp through my college’s Center for Career Services and alumni network. A Senior Account Executive at DataCamp, <a href="https://www.linkedin.com/in/samseath/">Sam Seath</a> (HWS Class ‘14), had let HWS know about an opening on the Sales team based in New York. After a few interviews, it felt like a great fit, and I was offered a role as a Business Development Representative!</p>
<h2 id="describe-your-current-team-what-do-you-find-most-interesting-about-your-role-">Describe your current team. What do you find most interesting about your role?</h2>
<p>I moved from the Sales team to the Customer Success team in January 2019. Our CS team is made up of thoughtful, engaging, curious, and most importantly, entertaining people. Each member of the team is a pleasure to work with. We are constantly learning from one another, picking up new ideas, and continuously pushing each other to improve in various aspects of our roles. I’ve really learned a lot working with this crew and I’m looking forward to what the future holds for the DataCamp Customer Success team.</p>
<p>What I find most interesting about my role would be the wide range of people and personalities I find myself working with. I can be on a call with an intern at a company who wants to learn more about DataCamp, and then 30 minutes later, I can be having a conversation with a Chief Data Officer of a Fortune 500 company about creating a training program for thousands of analysts. It just goes to show how many people in various roles and different company sizes can utilize DataCamp to improve their skill sets and grow their data skills.</p>
<h2 id="how-has-your-career-grown-since-joining-datacamp-">How has your career grown since joining DataCamp?</h2>
<p>When I started out at DataCamp, I was a Business Development Representative on the Sales team, sending hundreds of emails and calls a week to engage our user base and learn more about how companies were leveraging our platform. The goal was to open up opportunities for us to work with companies whose employees were actively using DataCamp. After about 6 months, <a href="https://www.linkedin.com/in/adamkuznia/">Adam Kuznia</a> (who was the head of CS and is now our CRO) thought I would be a good fit for a new role: Customer Success Representative (CSR). The CSR’s role is to engage our business clients to support their teams and help them to successfully leverage everything DataCamp has to offer. I also work with organizations to expand access to DataCamp for larger groups of their employees or for the entire company. This past May, I was promoted to Customer Success Manager, which means the accounts I now manage are much larger, covering hundreds or even thousands of employees. My clients represent various industries ranging from Finance, Healthcare, Tech, Government, Manufacturing, Transportation, and more. I enjoy helping them to successfully reach their goals with DataCamp.</p>
<h2 id="what-are-you-most-excited-for-in-2020-">What are you most excited for in 2020?</h2>
<p>Hopefully a return to some sort of normalcy! As nice as it is working remotely, I do miss the views from our Empire State Building office, and I miss collaborating with teammates in person, especially when I have quick questions or comments. Also, I’m really looking forward to/keeping my fingers crossed for the return of the NCAA and NFL Football seasons.</p>
<p>Besides the obvious, I’m looking forward to seeing my clients succeed on DataCamp. This year, I’ve worked with quite a few customers to align on training programs and paths that are specific to the needs of hundreds or even thousands of employees. I’m excited to see how these programs will come to fruition and it’s really great to know that we play a major role in spreading data fluency across some of the largest companies in the world!</p>
<h2 id="what-do-you-like-best-about-working-for-datacamp-">What do you like best about working for DataCamp?</h2>
<p>My favorite thing about working for DataCamp is the freedom that DataCamp provides to its employees. As long as you accomplish what you need to get done in a timely manner, nobody is going to ask questions. DataCamp allows us to balance work and life in a way that suits the individual. That being said, in a client-facing role, requests are coming in at all hours of the day, so it helps that DataCamp allows us to work on our own schedules without being stuck in a set 9 to 5 schedule.</p>
<h2 id="what-is-favorite-datacamp-cultural-value-and-why-">What is favorite DataCamp cultural value and why?</h2>
<p>“Put the team first.” When everyone adapts that team-first mentality, we all win. Whether it’s bouncing ideas off one another, inviting team members from other departments to join calls, or just having conversations at lunch, we’re always trying to support one another and help each other improve in any way possible. Currently, DataCamp is at a point where we’re growing rapidly and there are big things on the horizon. If we all share a common goal and continue to put the team first, there’s no limit to where this company can go.</p>
<h2 id="if-you-could-have-anyone-as-your-mentor-who-would-it-be-">If you could have anyone as your mentor, who would it be?</h2>
<p>I would have to go with Welles Crowther. Also referred to as “The Man in the Red Bandana,” Welles Crowther was an equities trader in New York who worked in the World Trade Center during 9/11. When the attacks happened, he could’ve made it out with his life, but he was a hero and went back in and helped rescue at least 18 others from the building. He sacrificed his life in order to save others, and that’s the type of person I would want as my mentor. I would highly recommend people learn about his story and watch a great documentary about his story: <a href="https://www.maninredbandana.com/">Man in Red Bandana</a>.</p>
<h2 id="why-do-people-call-you-mac-">Why do people call you Mac?</h2>
<p>My full name is James McInerney Stephens. Since we already had several Jameses in our family, they had to mix it up somehow. Ever since I was a baby, everyone I know has called me Mac. It’s kind of funny at this point—all of my family and friends refer to me as Mac, but all of my clients call me James. At this point it’s pretty interchangeable!</p>
<p><em>Would you like to join Mac at DataCamp? We’re currently hiring in New York; London; Leuven, Belgium; and remote locations. Check out our <a href="https://www.datacamp.com/jobs/">open roles</a>.</em></p>


https://www.datacamp.com/community/blog/james

Q&A with James (Mac) Stephens, Customer Success Manager at DataCamp

Employee Spotlight: Putting the Team and the Customer First

<p><center><a href="https://www.prnewswire.com/news-releases/datacamp-partners-with-anaconda-to-enable-data-fluency-for-teams-301078884.html"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1602873042/Anaconda_inklkv.png" /></a></center></p>
<p>The DataCamp community is growing. We are thrilled to announce our partnership with the popular open-source, cross-platform tool Anaconda to empower everyone with the confidence to code and become more data fluent. <a href="https://attendee.gotowebinar.com/register/8530113870488977166?source=blog">Join our webinar</a> on July 9 at 12 PM EDT to learn how to get started with Anaconda.</p>
<blockquote>
<p>Anaconda is proud to partner with DataCamp, a leader in online courses for data scientists. With a wide variety of courses from beginner to intermediate, DataCamp is a great resource for Python and R users.—Angela Pierce, President and CFO at Anaconda</p>
</blockquote>
<h2 id="datacamp-anaconda-team-edition">DataCamp + Anaconda Team Edition</h2>
<p>Through Anaconda Team Edition, members have access to more than 7,500 powerful open-source packages for data science and machine learning, alongside tools to ensure data science teams can use it securely. Anaconda’s mirrored repository is easily installed inside your firewall in front of your database management system.</p>
<p>Now, Anaconda Team Edition members will also have access to DataCamp’s expert data science training, including <a href="https://www.datacamp.com/courses/conda-essentials">Conda Essentials</a> and <a href="https://www.datacamp.com/courses/building-and-distributing-packages-with-conda">Building and Distributing Packages with Conda</a> via our <a href="https://www.datacamp.com/skill-tracks/anaconda-skills">Anaconda Skills</a> track.</p>
<h2 id="get-started-with-datacamp-anaconda-team-edition">Get started with DataCamp + Anaconda Team Edition</h2>
<p>Sign up for our webinar on <a href="https://attendee.gotowebinar.com/register/8530113870488977166?source=blog">Getting started with Anaconda</a> on July 9 at 12 PM EDT.  Even if you can’t attend live, we encourage you to sign up so you’ll receive a link to the recording.</p>
<p>For more information, <a href="https://www.prnewswire.com/news-releases/datacamp-partners-with-anaconda-to-enable-data-fluency-for-teams-301078884.html">read our press release</a> on the Anaconda partnership and <a href="https://gateway.on24.com/wcc/eh/2336732/lp/2375650/building-data-fluency-in-an-organization">watch our presentation</a> on building data fluency in an organization at AnacondaCON 2020. If you’re an analytics professional, you can explore our <a href="https://www.datacamp.com/learn-python-with-anaconda/">courses developed in partnership with Anaconda</a>.</p>


https://www.datacamp.com/community/blog/anaconda

DataCamp is partnering with Anaconda to help you develop your team’s expertise in Python for data science.

 DataCamp’s New Partnership With Anaconda to Spread Data Fluency 

<p><center><a href="https://www.datacamp.com/discover/enterprise"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591902844/4_steps_to_building_a_successful_data_program_o5wdgu.pdf" /></a></center></p>
<p><center><a href="https://www.datacamp.com/groups/business"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1603718763/banner_style_1_cruuf7.png" /></a><br></p>
<p>It can be challenging getting a data program off the ground, but without one, your business may not be able to synthesize important data to draw meaningful insights. Our recent DCVirtual conference dove into how to implement and measure a successful data-driven culture. DataCamp’s Curriculum Architect, Richie Cotton, spoke with AXA XL’s Head of Strategic Design, Data, Pricing and Analytics, Rachel Alt-Simmons, on this topic. Read on for some useful takeaways on why everyone needs to be data-driven, how to determine your data objectives, and how to implement a data program.</p>
<h2 id="it-s-everyone-s-responsibility-to-be-data-driven">It’s everyone’s responsibility to be data-driven</h2>
<p>In the 21st century, every organization needs to be data-driven to be competitive. Being data-driven is not about automating people out of their jobs, but providing them with the tools and information to be more effective in their roles.</p>
<blockquote>
<p>We think of it as everyone&#39;s responsibility in the organization to be more data-driven. After all, every single one of us is probably touching data in some way, regardless of your role. We&#39;re continually looking at different opportunities we have to upscale resources and get them along in the journey.—Rachel Alt-Simmons, AXA XL</p>
</blockquote>
<h2 id="data-program-success-metrics-depend-on-your-industry-and-organization">Data program success metrics depend on your industry and organization</h2>
<p>There is no one-size-fits-all data program. Each data program’s objectives will vary depending on factors like industry, company size, and team structure.</p>
<p>Let’s use AXA XL as an example. AXA XL, acquired by AXA Group in 2018, is a large commercial and specialty insurance company that assesses complex ventures. They currently employ 13,000 people around the world working on distributed teams, which means many of their teams don’t actually work with most of their team members. Virtual collaboration is important across global locations, and as an insurance company, their data needs are highly reliant on paper and PDFs.</p>
<blockquote>
<p>Our business runs on paper. We sell legal documents—that&#39;s what an insurance policy is, just a legal document. We intake paper and we generate paper. What happens is, for example, you get a submission in from a client, it contains tons and tons of files and information. You may get that information in a PDF, in Word documents, in Excel files, and probably every variation thereof. There&#39;s no standardization in how that document comes in, so there&#39;s not one standard PDF or one standard file. It&#39;s all different for every single client.</p>
</blockquote>
<p>Because so much of their data is stored in PDFs and data entry is manual, not all the individual data elements are captured. A lot of information is lost or not easily retrievable. To address these issues, AXA XL leverages capabilities like natural language processing to grab documents and easily scrape information out of them.</p>
<h2 id="how-to-implement-a-data-program">How to implement a data program</h2>
<h3 id="1-choose-a-top-down-or-bottom-up-approach">1. Choose a top-down or bottom-up approach</h3>
<p>The top-down approach to building a data program is related to the larger data and analytics initiatives within an organization. This requires strategically building out your program to standardize how analytic models and workflows are integrated across teams.</p>
<p>The bottom-up approach is about generally raising the capabilities and skill levels of all employees in a data and analytics career path. For example, AXA XL’s Analytic DNA initiative identifies experts in different technology and analytics areas, and looks to them for ideas to leverage to scale more broadly across the organization.</p>
<blockquote>
<p>The great news is that people have been really willing to step up and provide their guidance and expertise, and it helps us shape our program into something that&#39;s relevant and aligns to our strategic initiatives.</p>
</blockquote>
<h3 id="2-develop-personas-and-identify-their-goals">2. Develop personas and identify their goals</h3>
<p>At DataCamp, we’ve identified several data-related personas or roles that our customers typically use: Data Consumer, Leader, Data Analyst, Citizen Data Scientist, Data Scientist, Data Engineer, Database Administrator, Statistician, Machine Learning Scientist, and Programmer. We advise our instructors to consider who the relevant personas are when building their courses, and we advise our business customers to consider who the relevant personas are when building out their data programs. Each persona has a different relationship with data.</p>
<p>Two primary personas for AXA XL are Actuaries and Data Scientists, and they’re currently focused on converging the skills required of both roles to further the business.</p>
<blockquote>
<p>There&#39;s always been an interesting divide in many [insurance] organizations around the difference between an actuarial background and what I would call more of a pure data science or analytic background. In our organization, we&#39;re bringing those two things together. We&#39;re actually teaching actuaries some of those more statistical machine-learning techniques that they can apply in their actuarial processes. It&#39;s not an “us versus them” anymore. It&#39;s “how do we bring these capabilities together,” which is pretty exciting.</p>
</blockquote>
<h3 id="3-identify-gaps-for-each-persona">3. Identify gaps for each persona</h3>
<p>We advise all of our customers to take a broad inventory of their in-house data skills using a skills matrix. This comes in many different forms, but generally involves visualizing strengths and skill gaps at either the organization, department, team, or individual level.</p>
<p>Rachel at AXA XL advises that you can identify gaps for each persona by asking questions like:</p>
<ul>
<li>What is it that you want to know?</li>
<li>What would you like more of?</li>
<li>What more could we provide, e.g., training, hiring, skill development, or project-based work?</li>
<li>Are we aligning these capabilities with our strategic projects correctly?</li>
</ul>
<p>For analytics professionals like data analysts, data scientists, and data engineers, companies may wish to investigate whether there are gaps in important skills like importing data, conducting data analysis in spreadsheets, natural language processing, and building and maintaining data pipelines. <a href="https://s3.amazonaws.com/assets.datacamp.com/email/other/SkillMatrix.pdf">Here is a sample skill matrix that we developed for data professionals.</a></p>
<p>For Data Consumers and Leaders, skills needed for success include soft skills like communication, <a href="https://www.datacamp.com/courses/data-visualization-for-everyone">understanding data visualization</a>, and design thinking.</p>
<blockquote>
<p>Design thinking is [about] creating a solution, whatever it is: technical, data, [or] model. How do you do that in a way that&#39;s relevant for the customer? Then, how do you communicate the results effectively?</p>
</blockquote>
<h3 id="4-create-learning-journeys-for-each-persona">4. Create learning journeys for each persona</h3>
<p>A holistic learning approach requires using design thinking to create an effective learning journey. It isn’t only about building technical skills—but also about building important business skills. For instance, AXA XL requires their actuaries to be able to translate data into meaningful terms so that decision makers like underwriters, who evaluate insurance risk, can put a company’s risk assessment into context and draw useful conclusions, like how much risk to assume. The actuaries’ learning journey must include the ability to contextualize data and translate their implications to a broader audience.</p>
<p>To help organizations discover their team’s data skill level across personas, we offer a free <a href="https://skill-audit.datacamp.com/discover-your-companys-data-skill-level">skill audit</a> via a benchmark report. Individuals may take our <a href="http://datacamp.com/signal">Signal skill assessments</a> to track their own progress.</p>
<p>After assessing skills, we advise organizations to build custom tracks so that everyone can achieve the baseline skill level required for their persona or role. Learning tracks for Data Consumers and Leaders may skew toward our <a href="http://datacamp.com/courses/tech:theory">theory</a> courses, while data professionals take our hands-on coding courses. For instance, since so much of AXA XL’s data is in PDFs, learning journeys for their actuaries and data scientists include several of our courses on <a href="https://datacamp.com/search?q=natural%20language%20processing">natural language processing</a>. For early-career actuaries, we recommend these <a href="https://s3.amazonaws.com/assets.datacamp.com/email/other/Insurance+Courses+for+Predictive+Analytics+Exam.pdf">R courses to pass the Predictive Analytics Exam</a>.</p>
<h2 id="reap-the-benefits-of-your-data-program">Reap the benefits of your data program</h2>
<p>Rachel says it’s easy to measure the success of your data program by answering the question: <strong>Has it helped you become more profitable?</strong> If data is properly feeding into decisions, then your bottom line will reflect that. So if your employees’ skills are evolving and your company is making informed decisions, then the likely answer is: yes.</p>
<p>For more about building a successful data program, <a href="https://www.datacamp.com/resources/webinars/success-metrics-for-your-data-program">watch the webinar on-demand</a>.</p>
<p><center><a href="https://www.datacamp.com/business/demo"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1603718763/banner_style_2_m573ef.png" /></a></p>


https://www.datacamp.com/community/blog/success-metrics-for-your-data-program

AXA XL’s Head of Strategic Design, Data, Pricing and Analytics breaks down how to get your data program off the ground and implement a successful data-driven culture.

4 Steps to Building a Successful Data Program

<p><center><a href="https://www.datacamp.com/discover/enterprise"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1602862307/Adel_Spotlight_xfzqci.png" /></a></center><br></p>
<h2 id="what-was-your-journey-to-joining-datacamp-">What was your journey to joining DataCamp?</h2>
<p>Ever since I could remember, I’ve always been really curious about how different fields can intersect to give us a clearer picture of the world and the times we live in. So in some sense, I consider myself to be a bit of a polymath, and my path to joining DataCamp was a bit off the beaten path. I was born and raised in Beirut, Lebanon and I graduated with a bachelor&#39;s degree in Economics from the American University of Beirut.</p>
<p>After graduation, I was torn between pursuing a career in technology or political science. I managed to combine a bit of both as a research assistant intern at a think tank by scraping Twitter data to supplement a few research papers on Middle Eastern politics. Working on that project helped me to realize that working with data was something I was passionate about and that the implications of adopting data science at scale had massive implications for the societies we live in.</p>
<p>This is why I decided to pack my bags and move to Paris for a master’s program in Data Science. At grad school, I discovered DataCamp and it was such a valuable learning tool that it became an integral part of my data science education (and it also helped me pass a few courses 😉). One of my favorite things as an aspiring data scientist was demystifying complex tools and algorithms used in data science to non-technical audiences, so joining DataCamp&#39;s content team was a natural fit.</p>
<h2 id="describe-your-current-team-what-do-you-find-most-interesting-about-your-role-">Describe your current team. What do you find most interesting about your role?</h2>
<p>As a Content Developer on the Content team at DataCamp, my role is to work with our wonderful instructor community to develop <a href="https://www.datacamp.com/courses">courses</a> and <a href="https://www.datacamp.com/discover/webinars">live training sessions</a> on R, Python, SQL, and more, as well as to develop my own courses and live training sessions.</p>
<p>The most interesting part of my role is really all the knowledge that I soak up from working with the best data scientists in the world, and the satisfaction of creating a great experience for millions of learners by creating the best content possible.</p>
<h2 id="how-has-your-career-grown-since-joining-datacamp-">How has your career grown since joining DataCamp?</h2>
<p>Since joining DataCamp, my career has grown immensely. There are always opportunities to grow my skills and take on more and more challenging projects, all while having access to an incredible array of mentorship from DataCampers on the content team and beyond.  </p>
<h2 id="what-are-you-most-excited-for-in-2020-">What are you most excited for in 2020?</h2>
<p>Albeit 2020 has not given us much to be hopeful about outside DataCamp, I&#39;m really excited about 2020 and DataCamp&#39;s commitment to spreading data fluency across organizations. This commitment really drives our willingness to continuously improve on our learning experience, as well as innovate in the type of content we deliver, like live training sessions (watch this space 😉).</p>
<h2 id="what-do-you-like-best-about-working-for-datacamp-">What do you like best about working for DataCamp?</h2>
<p>Without a doubt, the best thing about working at DataCamp is the people. I&#39;ve had some of the most riveting conversations of my life with other DataCampers—and I&#39;d be hard-pressed to find a culture elsewhere that values care and empathy for other team members as much as DataCamp does. I also feel very privileged to work on a product that consistently improves people&#39;s lives by enabling them with the skills to compete in today&#39;s world at a fraction of the price of a college education.</p>
<h2 id="what-are-your-favorite-datacamp-cultural-values-and-why-">What are your favorite DataCamp cultural values and why?</h2>
<p>One of my favorite DataCamp cultural values that I alluded to earlier is care for other DataCampers and putting the team first. We really try to be an organization that does not tolerate a non-team based approach, while also cultivating a sense of empathy and care for everyone on the team.</p>
<h2 id="what-s-the-most-interesting-or-most-important-trend-you-see-today-">What’s the most interesting or most important trend you see today?</h2>
<p>Apart from the political turmoil we see today, I think one of the most important trends/stories we see today is really the continuous rise of surveillance technology and how it intersects with how economic value is defined today. <a href="https://www.google.com/books/edition/The_Age_of_Surveillance_Capitalism/lRqrDQAAQBAJ?hl=en&amp;gbpv=1&amp;printsec=frontcover">The Age of Surveillance Capitalism</a> by Shoshana Zuboff is a great book on this matter.</p>
<h2 id="anything-else-you-d-like-to-share-with-our-learners-">Anything else you’d like to share with our learners?</h2>
<p><a href="https://www.datacamp.com/discover/webinars">Join our live trainings!</a></p>
<p><em>Would you like to join Adel at DataCamp? We’re currently hiring in New York; London; Leuven, Belgium; and remote locations. Check out our <a href="https://www.datacamp.com/jobs/">open roles</a>.</em></p>


https://www.datacamp.com/community/blog/adel

Q&A with Adel Nehme, Data Evangelist at DataCamp

 Employee Spotlight: Creating Data Science Content for Millions of Learners

<p><center><a href="https://www.datacamp.com/discover/enterprise"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591190691/9_new_world_languages_2_ah732j.jpg" /></a></center><br></p>
<h2 id="datacamp-learners-across-the-world-are-turning-on-subtitles-in-their-native-languages-">DataCamp learners across the world are turning on subtitles in their native languages.</h2>
<p>Our mission is to democratize data science and analytics education for everyone. But until recently, we only had subtitles in English, the global language of business. The problem? Many of our learners are not native English speakers and <a href="https://support.datacamp.com/hc/en-us/requests/new?ticket_form_id=360000029413">gave us feedback</a> that they prefer to learn in their native tongue. To support our learners around the world, we developed course subtitles in nine popular languages: <strong>Spanish, French, Chinese [simplified], Portuguese, German, Korean, Japanese, Russian,</strong> and <strong>Italian</strong>. <a href="https://support.datacamp.com/hc/en-us/articles/360007678574-Are-DataCamp-s-video-subtitles-available-in-multiple-languages-">Find out how to change subtitle languages in our help center.</a></p>
<p>We already knew subtitles were popular. About <strong>30% of learners that start one of our courses enable subtitles</strong>—yours truly included. Combining visual and auditory learning styles can be effective. Many learners want the option to read along as they listen, which can help with comprehension.  Further, subtitles make our catalog much more accessible to people of different backgrounds and needs.</p>
<p>We don’t want our learners who aren’t native English speakers to have to work twice as hard to achieve the same goal as others. We want everyone to be able to understand and communicate in the language of data.</p>
<p>We launched the new subtitles in nine different languages for all our courses last month. Since then, we’ve seen that <strong>20% of our learners who turn on subtitles use non-English subtitles,</strong> with Spanish, Chinese, and Portuguese leading the way.</p>
<p>Our Product team would never have embarked on this journey without the input from our learners that they wanted to learn in different languages. Please continue to let us know <a href="https://support.datacamp.com/hc/en-us/requests/new?ticket_form_id=360000029413">how we can improve product features</a> like our translations and recommend which languages we should consider adding next. We would love to hear your <a href="https://support.datacamp.com/hc/en-us/requests/new?ticket_form_id=360000029453">feedback on our courses</a> as well! Together, we can make DataCamp a place where people from all walks of life can learn the language of data.</p>


https://www.datacamp.com/community/blog/subtitle-translations

Data science knows no bounds, which is why our video subtitles are now available in 9 new languages in addition to English: Spanish, French, Chinese (simplified), Portuguese, German, Korean, Japanese, Russian, and Italian!

Learn in Your Native Language With Our New Subtitle Translations

<p><center><a href="https://www.datacamp.com/groups/business"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591190738/Don_t_Make_Arrogant_Models_dafvec.jpg" /></a></center><br></p>
<h2 id="arrogance-is-not-a-good-quality-for-your-models-">Arrogance is not a good quality for your models.</h2>
<p>It’s a rarely acknowledged fact that models data scientists produce are often not sufficiently robust or fault-tolerant to actually be put into production. Sure, you can trust your predictions when the input data is similar to your training and holdout data. Any data scientist can build a model object, pass in data in the same format as the training data, and get seemingly reliable predictions.</p>
<p>In the real world, things are always messier than you expect. When data scientists throw models over the fence, ML engineers or the IT department often have to rebuild the models with guardrails to ensure they’re useful in production. As data scientists, we should do better—and do away with the arrogance of assuming our models will function as we intend in every production scenario. We should never say, “Here’s my Jupyter notebook; my work is done!” At the very least, we should start by documenting the expected behavior of incoming variables to help ML engineers write runtime tests more easily. Even better, I recommend adding a layer of unit testing to adjust model predictions.</p>
<p>In this article, I will walk you through a simple error-handling example using R with logical conditions wrapping a model’s predict function.  Python users are likely familiar with <code>assert</code>, <code>try except</code> and the usual logical operators to accomplish many of the same things covered in this article. This article uses R as an example because many data scientists using R don’t expose their models as endpoints for production, which means these model “humility” aspects may be new to them.</p>
<p>In the end, the goal is to add a layer of protection to your model to enforce expected behaviors so that it can withstand outliers, be fault-tolerant, and in some cases, override a prediction to a safe value.
These runtime tests and coded guardrails help make models as safe for production as a POST request within an opencpu server. Including these additional prediction behaviors within your model function will build trust among stakeholders that your model isn’t behaving arrogantly, and will deliver value despite outlier or unexpected inputs.</p>
<h2 id="example-set-up">Example set-up</h2>
<p>In this example, you will <a href="https://s3.amazonaws.com/assets.datacamp.com/email/other/final_Small_Customer_Data.csv">use this small sample dataset</a> to build a customer propensity model. This fake data has the results of a marketing campaign for a car loan offer. The input variables include current car make and recent savings account balance. Our classification model will learn which car makes and account balances contribute to accepting the marketing offer. Of course, in the real world you would have more data and follow stricter data science practices like partitioning—but in this example, we’ll take some shortcuts since we’re focusing on the prediction layer.</p>
<p>The [rpart] library is used for recursive partitioning to construct our decision tree. Similarly the <a href="https://www.rdocumentation.org/packages/rpart.plot"><code>rpart.plot</code></a> library will help us quickly construct a decent-looking visual of our tree. Next we use <a href="https://www.rdocumentation.org/packages/yardstick"><code>yardstick</code></a> to easily get model metrics and <a href="https://www.rdocumentation.org/packages/ggplot2"><code>ggplot2</code></a> to construct a mosaic plot.  The code below simply loads the data with <a href="https://www.rdocumentation.org/packages/utils/versions/3.6.2/topics/read.table"><code>read.csv()</code></a> and examines the first six rows with <a href="https://www.rdocumentation.org/packages/utils/versions/3.6.2/topics/head"><code>head()</code></a> so you get a sense of the inputs.</p>
<pre><code class="lang-r"># Libs
library(rpart)
library(rpart.plot)
library(yardstick)
library(ggplot2)

# Read in the data
fakeCustomers &lt;- read.csv(&#39;final_Small_Customer_Data.csv&#39;)

# EDA
head(fakeCustomers)
</code></pre>
<h2 id="let-s-build-a-simple-decision-tree">Let’s build a simple decision tree</h2>
<p>Now we apply the <a href="https://www.rdocumentation.org/packages/rpart/versions/4.1-15/topics/rpart"><code>rpart()</code></a> function to construct our decision tree. Since you’re accepting all default model parameters, you only need to pass in the model formula Y_AcceptedOffer ~ . and the data to make the tree. However, using the period (<code>Y_AcceptedOffer~.,</code>) in the model formula adds risk to your model’s behavior.  Suppose later the underlying training data changes to include additional columns.  By using the period the model will simply inherit all columns not defined as the Y-variable.  So if you rebuild the model by sourcing this code with data that has changed without explicitly declaring the x-variables, you invite target leakage or overfitting without even knowing it.  Thus, its often a good idea to declare x-variables explicitly in the formula.  In the end, the resulting <code>fit</code> object is a model which we do not want to simply pass to the IT department. Let’s also define a safe model response when fit gets an unknown value!</p>
<pre><code class="lang-r"># Fit the model
fit &lt;- rpart(Y_AcceptedOffer~., fakeCustomers)
</code></pre>
<h2 id="make-some-predictions">Make some predictions</h2>
<p>Let’s make sure our model functions as expected with perfect inputs. At this point in your workflow, you should be assessing model performance against a training and validation set. Here we use <a href="https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/predict"><code>predict()</code></a> on the original data, examine a portion of it with <a href="https://www.rdocumentation.org/packages/utils/versions/3.6.2/topics/head"><code>tail()</code></a>, then construct a simple confusion matrix.  Finally, you create a confusion matrix with <code>table()</code> and then nest <code>yardstick</code>’s <a href="https://www.rdocumentation.org/packages/yardstick/versions/0.0.6/topics/conf_mat"><code>conf_mat()</code></a> inside <code>summary()</code> to get 13 model metrics including accuracy.  Keep in mind marketers don’t have unlimited budgets, so you should care more about accuracy within the top 1% or 5% of prospects rather than just accuracy alone.</p>
<pre><code class="lang-r"># Get predictions
pred &lt;- predict(fit, fakeCustomers, type =&#39;class&#39;)

# Examine
results &lt;- data.frame(preds   = pred,
                      actuals = fakeCustomers$Y_AcceptedOffer)
tail(results, 10)
</code></pre>
<pre><code class="lang-r"># Simple Confusion Matrix
(confMat &lt;-table(results$preds, results$actuals))
</code></pre>
<pre><code class="lang-r"># Obtain model metrics
summary(conf_mat(confMat))
</code></pre>
<h2 id="visual-inspection">Visual Inspection</h2>
<p>In addition to numeric KPI you can inspect the confusion matrix visually with a mosaic plot.  In this example, a mosaic plot will have rectangles representing each section of the confusion matrix such as true positives and false positives.  The area of each rectangle corresponds to the value from the confusion matrix.  This view lets you easily understand how balanced your class assignments are compared to actuals.  The code below nests the original confusion matrix in <a href="https://www.rdocumentation.org/packages/yardstick/versions/0.0.6/topics/conf_mat">conf_mat</a> and ggplot2’s <a href="https://www.rdocumentation.org/packages/ggplot2/versions/3.3.0/topics/autoplot">autoplot</a> function to create a basic mosaic plot.</p>
<pre><code class="lang-r">autoplot(conf_mat(confMat))
</code></pre>
<p>One benefit of using a simple model is that you can interrogate the model’s behavior.  For decision trees you can use <a href="https://www.rdocumentation.org/packages/rpart.plot/versions/3.0.8/topics/rpart.plot">rpart.plot()</a> function to visualize the result. This plot will let you understand the split values and importance of variables in each node.</p>
<pre><code class="lang-r">rpart.plot(fit, roundint = F)
</code></pre>
<h2 id="all-good-right-not-so-fast-">All good right? Not so fast.</h2>
<p>Don’t send this model code to IT and expect a warm response! Sure, it works just fine with these fake prospects—because they are exactly like the training data. Even in normal model building, you usually pass in a partition with similar distributions and certainly the same factor levels. But in reality, data integrity and other factors can be issues with the real incoming data—<strong>and they can break your model.</strong></p>
<pre><code class="lang-r">fakeNew &lt;- fakeCustomers[c(6:8),]
fakeNew

# Make a prediction
predict(fit, fakeNew, type = &#39;prob&#39;)
</code></pre>
<h2 id="add-a-layer-of-protection-for-your-predictions-">Add a layer of protection for your predictions.</h2>
<p>In this section, you explore what happens when the make of a car is changed from Lexus to lexus. Data entry errors and mis-keys happen all the time because people are involved. Mis-keying factors and transposing numerical inputs often break models in production, as you’ll see if you run predict(fit, fakeRecord) below.</p>
<p>Error: factor carMake has new level lexus</p>
<pre><code class="lang-r"># Entry Form Error
fakeRecord &lt;- fakeNew[1,]
fakeRecord[,2] &lt;- as.factor(&#39;lexus&#39;)

# Uh-Oh!; Error: factor carMake has new level lexus
#predict(fit, fakeRecord)
</code></pre>
<h2 id="adding-a-humble-layer-to-your-model">Adding a humble layer to your model</h2>
<p>Let’s add a protective prediction layer by checking the inputs all make sense and if so then call <a href="https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/predict">predict()</a>. In this code, you write a wrapper function called humblePredict() that accepts the new observations to be scored.  Within a for-loop the function checks:</p>
<ol>
<li>That each row is part of a dataframe using <code>is.data.frame</code></li>
<li>That the dataframe’s columns match the model training formula using the match <code>%in%</code> operator.</li>
<li>That the value of the observation in the <code>carMake</code> column is an expected level from model training data.  This is another match operator call with <code>%in%</code></li>
<li>Finally, that the <code>&#39;RecentBalance&#39;</code> column is a numeric value using the <code>is.numeric</code> function.</li>
</ol>
<p>If all four of these logical conditions are met then the if statement simply calls <a href="https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/predict"><code>predict()</code></a> as usual.  However the logical conditions occur within an if-else statement.  Therefore if any of these conditions returns a FALSE, then the false code block executes.  In this example the default response is the “safe” model response of “DidNotAccept”.  This level is safe because it means the company wouldn’t spend money marketing to this potential customer.  Of course in your own work, you could have a more explicit error, use a different model or simply return the average Y value from your training set.  The point is that you have complete control of the false code behavior and should ensure that your model has guardrails corresponding to the business need.  This type of function wrapping helps you dictate how your model should behave with bad inputs.  Do you want an error, a safe value, NA or other output when a model is confronted with bad inputs?</p>
<pre><code class="lang-r">humblePredict &lt;- function(x){
  classifications &lt;- list()
  for(i in 1:nrow(x)){
    if(is.data.frame(x[i,]) == T &amp;
     all(all.vars(formula(fit)[-2]) %in% names(x[i,])) == T &amp;
     x[i,grep(&#39;carMake&#39;,names(x[i,]))] %in%
     unlist(attributes(fit)$xlevels) == T &amp;
     is.numeric(x[i,grep(&#39;RecentBalance&#39;, names(x[i,]))])==T){
    response  &lt;- predict(fit, x, type = &#39;class&#39;)
    classifications[[i]] &lt;- response
  } else {
    response  &lt;- &#39;DidNotAccept&#39;
    classifications[[i]] &lt;- response
    }
  }

  return(unlist(classifications))
}
humblePredict(fakeRecord)
</code></pre>
<p>This is just the tip of the iceberg for making models more robust in production. You can write code within humblePredict to change outlier numeric inputs or change factor levels to the most frequent if a level is unknown. If you want to learn more, start with both the <code>testthat()</code> and <code>assertive()</code> libraries for unit testing and run-time testing respectively.  <strong>No model should be sent to IT without assertions or at least documentation for safe behaviors.</strong></p>
<p>Find out how to build AI you can trust with <a href="https://www.datarobot.com/">DataRobot</a>.</p>


https://www.datacamp.com/community/blog/ai-humility

How humility can help make your models more robust in production.

Don’t Make Arrogant Models

<p><center><a href="https://datacamp.com/courses/introduction-to-airflow-in-python"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591112596/blog_Intro_to_Airflow_lqqtkw.png" /></a></center><br></p>
<p>Apache Airflow is a crucial part of the data engineering ecosystem. That’s why our introductory data engineering courses, <a href="https://datacamp.com/courses/introduction-to-data-engineering">Introduction to Data Engineering</a>, <a href="https://datacamp.com/courses/building-data-engineering-pipelines-in-python">Building Data Engineering Pipelines in Python</a>, and <a href="https://datacamp.com/courses/data-engineering-for-everyone">Data Engineering for Everyone</a>, include lessons on Airflow. Now, we’re excited to announce the launch of our first dedicated course on Airflow: <a href="https://datacamp.com/courses/introduction-to-airflow-in-python">Introduction to Airflow in Python</a>.</p>
<h2 id="what-is-airflow-">What is Airflow?</h2>
<p>Apache Airflow is an open-source tool to create, monitor, and schedule workflows. In this context, a workflow is a set of steps to accomplish a data engineering task, like streaming data or writing data to a database. Airflow represents workflows as Directed Acyclic Graphs, or DAGs. Essentially this means workflows are represented by a set of tasks and dependencies between them, like this:</p>
<p><center><a href="https://datacamp.com/courses/introduction-to-airflow-in-python"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591116767/Airflow_Example_DAG_-_Clear_y8vgra.png" /></a></center><br></p>
<h2 id="who-s-this-course-for-">Who’s this course for?</h2>
<p>Are you responsible for delivering data on a schedule? Then this course is for you!
In terms of roles, this course is intended for data engineers, data scientists, and machine learning scientists who want to upskill in data engineering. The only prerequisites for this course are <a href="https://datacamp.com/courses/intermediate-python">Intermediate Python</a> and <a href="https://datacamp.com/courses/introduction-to-shell">Introduction to Shell</a>.</p>
<h2 id="what-you-ll-learn">What you’ll learn</h2>
<p>This is an introductory course where you’ll learn everything you need to get started with Airflow.</p>
<p>First, you’ll get comfortable with components of Apache Airflow such as DAGs and why they’re useful. Then, you’ll learn how to implement Airflow DAGs using operators, tasks, and scheduling. You’ll also learn about sensors and executors to monitor and debug Airflow workflows. In the final chapter, you’ll use what you’ve learned to build a production-quality workflow in Airflow.</p>
<h2 id="how-you-ll-learn">How you’ll learn</h2>
<p>This is one of the most interactive Airflow courses you’ll find online. You’ll use both the Airflow web UI, or user interface and the Airflow command-line interface (CLI) to schedule and monitor workflows. Along with classic coding exercises, this course has IDE (Integrated Development Environment) exercises, where you can look at multiple files and run different Airflow CLI commands in a real-world environment.</p>
<p><center><a href="https://datacamp.com/courses/introduction-to-airflow-in-python"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591117590/Screen_Shot_2020-06-01_at_12.46.30_PM_hr7xnj.png" /></a></center><br></p>
<p>To gain experience in different ways of interacting with Airflow, we incorporate exercises that embed the Airflow web UI:</p>
<p><center><a href="https://datacamp.com/courses/introduction-to-airflow-in-python"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1591116772/airflow3_wfiv1n.png" /></a></center><br></p>
<p>The instructor of this course, Mike Metzger, is a data engineer with 20+ years of experience in the data, networking, and security space. You may know him from his other fantastic DataCamp course, <a href="https://datacamp.com/courses/cleaning-data-with-pyspark">Cleaning Data with PySpark</a>.</p>
<p><strong>Enjoy the course and let us know what you think at <a href="mailto:support@datacamp.com">support@datacamp.com</a>!</strong></p>


https://www.datacamp.com/community/blog/airflow

If you’re responsible for delivering data on a schedule, learn how to manage your data engineering workflows with DataCamp’s first course devoted to Airflow.

New Course: Introduction to Airflow in Python

<center><a href="https://www.datacamp.com/discover/enterprise"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1590507220/AI_Finance_Algorithmic_Trading_adajcl.jpg" /></a></center>
<p><center><a href="https://www.datacamp.com/groups/business"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1603718763/banner_style_1_cruuf7.png" /></a><br></p>

<p>Recently, Dr. Hugo Bowne-Anderson (DataCamp) and Dr. Yves Hilpisch (The Python Quants) discussed how AI can be leveraged in finance, how it compares to more traditional approaches, and why data scientists are now key players in finance. They also discussed uncertainty and risk, and how useful AI and machine learning can be in market crashes, during black swan events, and in volatile markets. <a href="https://www.datacamp.com/resources/webinars/ai-finance-and-algorithmic-trading">Watch the webinar on-demand</a> or read on for some key takeaways.</p>

<h2>What is AI and what are the implications?</h2>
<p>Yves defines artificial intelligence as the whole range of fields that try to emulate and improve what people can do. In that sense, AI in chess tries to play chess better than humans, and AI in finance can mean building and training bots that can trade better than humans do.</p>
<blockquote>
<p>AI is where software achieves something that we would consider an intelligent act. And intelligence can be defined as the ability to achieve a certain goal.—Dr. Yves Hilpisch</p>
</blockquote>
<p>Machines can follow the scientific method much more efficiently than humans. They are able to look at the data and find out revealed preferences by tracking what people actually do. And from that, they derive insights. They leverage millions of data points to form these insights, which end up being far more accurate than theoretical models.</p>
<blockquote>
<p>So what is the scientific method?...You set up an experiment, you collect data, and then you try to either find support for the hypothesis, or maybe you find data that contradicts the hypothesis. Then the next step will be to adjust your hypothesis, your assumptions, or maybe retract the hypothesis altogether. This process of the scientific method can be replicated by machines a million times in seconds.—Dr. Yves Hilpisch</p>
</blockquote>
<h2>AI has provided a major competitive advantage in the past 20 years</h2>
<p>AI serves the same purpose in finance as it does for any other field: to gain a competitive advantage. In finance, it got its start in the 1950s, when Markowitz led the way with the first quantitative finance model. It was the first model in variance portfolio theory that gained widespread acceptance. Today, it’s still used to manage trillions of dollars. But it uses normative theory which is based on analyzing various possible portfolios of securities—it’s not prescriptive because it doesn’t suggest a possible course of action. So it’s not based on data, but rather behavioral data or market data.</p>

<p>In the past 20 years, data has become more readily available than ever before—the key is to use it appropriately. Gregory Zuckerman’s <a href="https://www.amazon.com/dp/B07NLFC63Y/ref=dp-kindle-redirect?_encoding=UTF8&amp;btkr=1">The Man Who Solved the Market</a> details how the hedge fund Renaissance Technologies led the way in amassing data for financial gain. Today, almost every industry has embraced their strategy of creating mathematical models and crunching data—and are able to make more optimal decisions as a result.</p>

<h2>Modern use cases and best practices for quantitative finance</h2>
<p>There has been a dramatic evolution of AI in financial markets. For example, using the mountains of data available today, supervised learning models are able to predict the behavior of creditors or consumers with a high degree of accuracy. Algorithmic trading also leverages reinforcement learning to reward and punish trading bots based on how much money they make or lose.</p>
<blockquote>
<p>We have now millions and millions of data points that we can use to analyze the behavior of people. And with the new technologies that we have, banks and institutions [such as] fintech startups are ten times, a hundred times better at predicting consumer behavior, creditor behavior, etc, than every theory that ever has been devised by financial professors.—Dr. Yves Hilpisch</p>
</blockquote>
<p>But in times like these where there is high market volatility, it’s clear that AI in finance requires safety precautions. Supervised learning relies on historical data and assumes that what has happened in the past is representative of what will happen in the future. This is faulty logic in times of market volatility. That’s why you must have rules in place to prevent bad trading decisions, like putting a stop loss—which specifies a price that a security or commodity will be sold—on the algorithms. This logic limits losses that are too great to bear and is akin to putting security mechanisms on a self-driving car to prevent it from hitting pedestrians on a sidewalk.</p>

<p>AI is useful for reducing risk resulting from current events and its impact on the market. In black swan events, supervised learning tactics must be carefully monitored, and reinforcement learning can be used to retrain models based on new market conditions. Human intervention and careful monitoring of deployed algorithms will still be required for the foreseeable future.</p>

<h2>The need for simple, replicable, scalable tools</h2>
<p>In computational finance, the Monte Carlo simulation has a forward-looking approach that starts with a certain point in time and simulates outward into the future. Python is a great tool for this.</p>
<p>The Python data stack today is a set of interoperable packages that allow you to ingest data, transform it with pandas, conduct statistical modeling, and create machine learning algorithms. Python allows all this to happen at scale.</p>

<p>Today, algorithmic trading depends on having the proper tooling. And getting a job in this industry requires a certain degree of proficiency in these tools—namely, Python and R. The Python package pandas and frameworks like TensorFlow and Keras have allowed many people to be able to do what was previously impossible.</p>

<h2>Lowering the barriers to entry</h2>
<p>Learning how to use these tools is a big investment, and more companies and individuals are making this investment than ever before. Educational offerings like DataCamp and <a href="https://home.tpq.io/">The Python Quants</a> can democratize the learning process. Learners become familiar with data science and machine learning concepts and get comfortable with building algorithms independently in a coding environment. They’re also able to get familiar with how to set up a proper environment and tool chain to work on the server—empowering them to deploy algorithms in the cloud.</p>
<br>
<p><center><a href="https://www.datacamp.com/business/demo"><img src="https://res.cloudinary.com/dyd911kmh/image/upload/f_auto,q_auto:best/v1603718763/banner_style_2_m573ef.png" /></a></p>


https://www.datacamp.com/community/blog/ai-finance-algorithmic-trading

As artificial intelligence, machine learning, and data science are fundamentally changing the world, we need to investigate how they’re impacting finance and algorithmic trading, particularly at a time when markets are so vulnerable and volatile. 

community

datacamp

Official Blog