A common question that comes up with Interactive HTML output is "Why is this report just an image?" We attempt to provide feedback in the Log to give a high-level indication of why. Not familiar with the log? You can access it in Windows by going to View > Log (or pressing CTRL+SHIFT+L), or on the Mac by going to to Window > Log (CMD + 1). The Log can include script output when working with JSL as well as informational output such as the messages output when exporting Interactive HTML. We avoid messages that are so vague that they frustrate power users and statisticians and messages that so specific that they sound like gibberish to less technical users. Hopefully, the messages are informative, and are useful to provide to Technical Support if you are requesting something be implemented in interactive HTML or reporting a problem. Below, I decode the most common causes for Profilers to come out as static images. Reason 1: Categorical response I mentioned this above, but currently only continuous responses are supported. If you have tried to model a categorical response and save to Interactive HTML or publish to JMP Live, you will get the message "Interactive HTML: Profiler with categorical response <column name> is not interactive" in the log. Reason 2: Unsupported JSL commands The JMP Profiler evaluates a mathematical formula with multiple factors, varying one while keeping the others fixed to produce a curve for each factor. Internally, the formula is a JSL expression, such as what you find when you view the properties of a column with a formula. JSL is a large language with many commands and operations. Some are too programatically intensive to perform in the browser, which might perform badly. Also, we are trying to implement the most widely used subset and will add features judiciously as we understand the demand for them from you, the user. In JMP 12, the same 50 or so commands implemented in Flash were ported: Add, Subtract, Multiply, Divide, Minus, Power, Root, Abs, Floor, Ceiling, Min, Max, Equals, NotEquals, GreaterThan, LessThan, GreaterThanOrEqualTo, LessThanOrEqualTo, Or, And, Not (negation), Exp, Log, Log10, Sine, Cosine, Tangent, SinH, CosH, TanH, ArcSine, ArcCosine, ArcTangent, ArcSinH, ArcCosH, OpArcTanH, If, Match, Choose, and a number of Financial calculations DDB, FinFV, FinIPMT, FinIRR, FinMIRR, FinNPER, FinNPV, FinPMT, FinPPMT, FinPV, FinRATE, FinSLN,FinSYD. In JMP 15, six additional ones were added (Is Missing, Zero or Missing, IfMZ, Modulo, Ln and Log1P), mostly because they were heavily used by the Partition, Boosted Tree and Bootstrap Forest platforms. There is no point in doing the work to enable a new embedded platform if the formulas it typically produces are not supported yet due to unimplemented JSL commands, so these had to be implemented first. If a command we haven't gotten to is encountered, the log will display the message "Interactive HTML: Profiler for <column name> is not interactive because of unsupported operations." Reason 3: Unsupported platforms If you go to Preferences > Platforms in JMP, you will see more than 100 platforms listed, many of them with Profilers.  Not counting the Prediction Profiler platform (found in Graph> Profiler), eight platforms are now supported. Again, we started with what we felt were the most widely used, and we want to hear from you if there is a platform you need that isn't supported. If your platform comes out static, look in the log and it will say "Interactive HTML: Embedded profilers in this platform are not interactive." If you get this message in the log, there may be a workaround. For many platforms that have a Profiler, the red triangle menu will provide a "Save Columns" submenu with "Save Prediction Formula" as an option: Demonstrating the Save Prediction Formula option in the Partial Least Squares platform. Using the Body Fat sample data set, run the "Partial Least Squares" script and add the Profiler from the red triangle menu for "NIPALS Fit with 7 Factors". If you were to export this, you would get a static image -- we haven't gotten to enhancing the Partial Least Squares platform yet to export the Profiler to Interactive HTML. However, if you go to that same red triangle menu and select Save Columns > Save Prediction Formula, it will create a new column "Pred Formula Percent body fat". Prediction column saved to the data table. By viewing the column's properties you can see the formula of the profiler. If you view the Column Properties, you can see there is nothing complex about this formula -- simple multiplication and addition. If you open up the Graph > Profiler menu and add this column for Y, you get a standalone Profiler that does export to Interactive HTML or JMP Live. Because of the number of factors, I chose to go to the red triangle menu and make use of the Arrange in Rows feature (found under the Appearance submenu) to make the Profiler wrap into two rows with a value of 7. Reason 4: Advanced features For individual embedded Profilers, there could be additional features that aren’t supported specific to that platform: Non-mixture linear constraints Random effects Multiple response variables Noise factors Offset variables (this is specific to Generalized Linear Model) If you encounter one of these, you will get a message in the log that should explain, for example: "Interactive HTML: Profiler with non-mixture linear constraint is not interactive." We continue to improve the output of the very popular Profiler feature in Interactive HTML. Hopefully, this blog post helps to explain some of the cases that we haven't implemented yet. We want to hear from you about the features and embedded platforms you would like to see -- leave me a comment! ... View more

This Profiler is embedded -- that is, displayed in the context of its platform report. JMP 15 brings the biggest enhancement to Interactive HTML Profilers in a while. We implemented new JSL methods in JavaScript that will enable more Profilers to come out interactive than ever before. In addition, four new embedded Profilers were enabled, more than any other release, as we strive to bring as much interactivity to JMP Live and JMP Public as possible with JMP 15. Besides a number of bug fixes, work was done to make additional embedded Profilers interactive in JMP 15. When I say "embedded Profilers," I mean a Profiler displayed in the context of its platform report, rather than the Prediction Profiler found under the Graph menu in JMP. As I've mentioned in previous posts, Fit Least Squares was done in first in JMP 12 since it is the most commonly used, and Neural, Generalized Regression and Generalized Linear Model were added in JMP 13. Four new platforms are supported in the newest versions of JMP and JMP Pro: LogLinear Variance, Partition, Boosted Trees and Bootstrap Forest. LogLinear Variance This personality of the Fit Model enables you to model both the expected value and the variance of a response using regression models. The following example from the Injection Molding sample data set runs the "Model" script in the data table, and then from the red triangle menu under the Profiler submenu, I select "Profiler". The output can be published to JMP Public or JMP Live using the Publish option under the File menu. The content below is hosted on JMP Public and embedded in this blog using an iframe.   Partition The Partition platform recursively partitions data according to a relationship between the predictors and response values, creating a decision tree. It is important to note that currently we still only support continuous responses where the platform fits the means of the response values; so while responses can be categorical in the JMP Partition platform, it is not yet implemented in Interactive HTML and will come out as an image within the larger interactive report. However, there are still plenty of continuous examples, such as this one taken from the diabetes data set.   Boosted Trees and Bootstrap Forest These two platforms are available only in JMP Pro, and are related to the Partition platform, making use of multiple decision trees. Head over to JMP Public to see a few examples of the Profilers in these platforms. ... View more

JMP is a powerful analytic and visualization tool, but in order to use it you need data. Many of the best features in JMP are related to importing your data – Query Builder, XML processing, multiple file import, and bringing in data with HTTP request are all newer features geared at bringing in data from other sources with ease so that you can analyze and visualize the data with JMP.  Recode and other features help you to take messy data and transform it into a clean data set, and the data table provides featues such as virtual join to combine data sources effectively.  JMP is great at handling your data ... assuming you have access to it. In our professions, we routinely work with data sources provided by clients and internal to our companies, analyzing data to increase profits, improve processes, and make decisions. As data scientists, some of us like to look at our own personal data, and this blog post began as I started to work on an upcoming paper about my running data. Mine was less of a personal story like a few of my colleagues have presented on the topic of quantified self and their own personal data; it was more about my interest in creating visualizations and finding predictors of success in a race by analyzing my training leading up to the race. However, creating an analytical paper like this does bring in aspects of my personal story, and I wanted to share my experiences in trying to gather my running data. I began running at NC State because I was required by my degree program to take an introductory PE class. Given the choices, I went with "Run Conditioning," which culminated in a 5K race. I got hooked on running and racing, and I purchased a Garmin Forerunner 301 that synced to a little Windows application that was great at viewing the data, but didn't provide visualizations and no way to share your info. At the time, I wasn't really interested in that, although I did like having a tally of my total miles run and weekly miles for motivation. Many runs and races are stored in various data sources over the years. In addition, I was carrying an iPod to listen to music, and later got an iPhone capable of acting as its own GPS. At the time, there wasn't a lot of competition, and Nike had developed Nike+, which had a way of handling running on the treadmill without GPS signal as well. I didn't want to have to put on a watch and carry my iPhone, which I was already doing to listen to music, so I moved over to Nike and retired the Forerunner. Garmin had a way of exporting to their TCX file format, and Nike at the time allowed importing that format, so I consolidated my data. Nike's website provided nice visualizations and Google-powered maps that I could access from anywhere I had the internet, a public API, the ability to import and export my data – and even easy ways to share runs to social media if I wanted to. This isn't a post about the alarming amount of data that people put into the cloud, social media, and third-party software or the security implications created by those decisions. I do find those topics important and interesting, and often overlooked, but I found myself a good 10 years later wanting to gather all of my data to work on this paper, and I couldn't. The Nike+ website no longer shows any of the visualizations and stats it once did – you have to use the latest iteration of their app called Nike Run Club. The app is fine and works much like the website used to, but there is no way to export data. If you search on the web, you find a graveyard of broken links to a now-removed REST API documentation, and a myriad of websites and apps created specifically to get your Nike data that have risen and fallen. Just about a year ago, I was able to extract my 2016 marathon run data using n+exporter website, which has since been broken by changes Nike made to its API. After much searching, I stumbled on mentions of an app called Run Gap, which works as a way to manage multiple data sources – you provide your login information, and it can import your data from more than 20 popular sources, including Nike for the time being. I downloaded it and gave it a try, and was finally able to extract my data, although in a very inconvenient format of about 700 individual GPX files. I may just pay a fee to unlock the ability in Run Gap to transfer data directly from one source to another. After discussing this issue with my colleague Julian Parris (@julian), I have moved over to logging my current runs in Strava. The API support and import/export capabilities are much more user-friendly (at the moment). My point is this: If you are relying on a third party to store your data, you could lose it at any time. It is spelled out in the fine print. The Nike Terms of Service  clearly state: NIKE may terminate or modify any Nike Platform, member program, product or service at any time without notice. NIKE may terminate or suspend your account, delete your profile or any of your User Content, and restrict your use of all or any part of the Platform at any time and for any reason, without any liability to Nike, subject to applicable law. This means Nike could for any reason delete my data or lock me out of my account. And this is not unheard of. For example, in 2015, LinkedIn decided to remove the ability to mass export your contacts, and there was a huge backlash. Luckily, LinkedIn listened to its users and reversed course. However, the importance of data to discovery is paramount – data-related challenges are one of the biggest hurdles to AI. Collecting data can be difficult when the data comes from other sources, but at least you can own your own data. So, go get your data while you can. ... View more