data scientists - opinions soughtWhat aspects of JMP/JMP Pro are/should be utilized, and most powerful by any self-proclaimed 'data scientists'?
Re: data scientists - opinions soughtTwo other thoughts...neither directly related to JMP products but I think these two types of skills can be a differentiator for a 'data scientist' from at least some of their colleagues.
Thought 1: IMO practice of data science related techniques (those in JMP products) is most efficient when the scientist has sound, testable, rational business questions or techical issues to resolve as presented by their consumers (not to mention appropriate data to answer these questions/issues). The situation where the 'data' shows up over the transom of the data scientist's boat, with the instructions limited to 'Tell us what it's telling us'...are a recipe for all sorts of problems, wasted effort, missed signals and less than optimal decisions. So when the data scientist is immersed in this situation, they need to be able to put the brakes on the analytics effort and ask their sponsors the probing questions so that the analytics piece of the process is executed efficiently and effectively.
Thought two: Consumers often don't care about the rich set of analytics tools that were used to answer the original questions or inform them. So the data scientist needs to be bilingual...speak the language of analytics (to defend or explain in detail their approach) and the language of their consumers. Usually that language is money or business impact...not R^2, F-tests, confidence intervals, and on and on. So if the data scientist only communicates in the language of analytics, there is a good chance the overarching message and their impact on the project will be less than fully effective. One way to help with the bilingual piece of this is use JMP's data visualization tools...not the tables of stuff that come with the platforms. The scientist will still probably have to explain some things...but if kept simple and visual...they have a fighting chance at being heard and understood.
Re: data scientists - opinions sought@dara_hammond Let me try and draw some lines in the sand...but recognize these are squiggly lines "Data Assembly" phase methods such as hand entry, JMP Query Builder, Query Builder, Exel/kissing cousin import flows, interface with SAS/R/Python/Matlab, etc. and now in version 14 multiple file import, http request. But even here...something like Query Builder could definitely get into 'Wrangling'. Like starting to filter, change variable names in the Query (just another form of Recode really?) and on and on.
In the 'Wrangling' space...I'll start with the presumption the data is now in a JMP data table. Now it's time to reshape, reformat, transform, create new variables (like a Validation column or use the formula editor) etc. So many data table transforms in the Tables menu, missing value detection/imputation, Explore Outliers for the 'how do I handle them?' question. And the mother of them all now in version 14, Recode.
Re: data scientists - opinions soughtAs I compile my thoughts aorund all the great ocmments I've gotten, I seem to be able to draw several lines in Peter's Phase 2 based on your separation of "Data Assembly" and "Wrangling". Can you share more detail on what you'd categorize where wrt the segmentation?
Re: data scientists - opinions soughtVery interesting article on using data to make sense of what "data science" means in the first place, from a job description point of view.
It might provide clues on how to help JMP customers making the transition in their careers, or to understand what new customer that self-identify as data scientists might be looking for.
https://medium.com/indeed-data-science/theres-no-such-thing-as-a-data-scientist-8dae923c14e3
Case in point: "Finally, we see that some natural scientists are perhaps getting into data science through the statistician end of the data science spectrum."
Re: data scientists - opinions soughtFor those that are interested I've attached the CRISP-DM reference document. As @nascif_jmp mentions it's dated...and SPSS was one of the contributing/participating vendors...but I think the authors did a pretty good job keeping the document vendor agnostic. I did some digging and it looks like the 2.0 version never got beyond some draft review stages so was, as near as I could tell, never officially published. But I think this document stands the test of time.
Re: data scientists - opinions soughtRe: analytics lifecycle, also worth looking up the CRISP-DM model. It goes beyond SAS and has been around for a while.
@russ_wolfinger wrote:
Yes, looks good. Fits in well with a lot of banter these days about the "analytics lifecycle"--part of the messaging of SAS Viya and also the topic of a book by Greg Nelson from Thotwave/Duke whom Anne just interviewed on Analytically Speaking.
Can add simulation to Phase I, and my preference would be to split Phase II into two, something like Data Assembly and Data Wrangling.
Also, keep an eye out for the "Data Science Pentagram" in my talk next week.
Re: data scientists - opinions soughtYes, looks good. Fits in well with a lot of banter these days about the "analytics lifecycle"--part of the messaging of SAS Viya and also the topic of a book by Greg Nelson from Thotwave/Duke whom Anne just interviewed on Analytically Speaking.
Can add simulation to Phase I, and my preference would be to split Phase II into two, something like Data Assembly and Data Wrangling.
Also, keep an eye out for the "Data Science Pentagram" in my talk next week.
Re: data scientists - opinions soughtAny capability that helps them solve the practical problem at hand. So I think the issue is really thinking about the workflow for any given problem. I see 4 potential phases to the use of JMP/JMP Pro/JMP Clinical/JMP Genomics...because I believe a data scientist, depending on the problem, might find any of these tool boxes optimal...again depending on the problem. I don't fully appreciate the tools in JMP Clinical and Genomics so my lists in the phases below certainly omit mission critical capabilities in those two products.
Phase I: Creating the data when none exists. Here I'm talking primarily about DOE as the means to create the data necessary for the problem at hand.
Phase II: Acquiring the data. Now the 'data' exists someplace else, and it's necessary to get the data into the JMP ecosystem. So Query Builder, direct file import, scraping from the web in some form, or even entering the data by hand might come into play. I'm going to include all the data preparation and vetting steps in this phase. So Recode, Explore Outliers, Missing Data, transformation/Formula Editor for creating derived variables, Validation tactics, etc.
Phase III: Analyzing the data. Good Lord...this could be the entire Analyze and Graph main menu bars, again, given the problem at hand.
Phase IV: Sharing your discoveries with others. Here I think it's important to think about the concept of 'producers' and 'consumers'. Producers create the analysis, conclusions, and recommendations, and answers to the original problem. Consumers need to either be informed, make a decision, or actually use the results in some fashion. Static things like copy/paste to MS Word/pdf, etc. might be sufficient for consumers that need to be informed. Interactive dashboards, journals, or HTML5 content might be needed for those that need to make a decision. And don't forget Formula Depot for publishing models to some destination production application.
Re: data scientists - opinions soughtData science is exploding and my sense is JMP has a very strong and unique play as a "first look" tool, especially for tabular data and increasingly so for text and image data. Got data tables? Open them in JMP first to quickly and very efficiently determine what you have, find problems (nearly every raw data set has them), wrangle, and engineer features for predictive modeling. By the way, you can also design optimal experiments to better understand and validate your systems and analyze results with best statistical models. While JMP's predictive modeling platforms are not state of the art, it's interactive exploratory capabilities and deep integration of graphics and analytics are. Mouse versus keyboard at first.
With kind nudgings from Dan V, Geoff, and others, I'm putting together a talk around lessons learned in the fires of Kaggle competitions and will be presenting March 22 at the JMP Dev meeting. JMP has been my secret weapon all along. Dara, thanks for the perfect trailer question; check is in the mail...
Re: data scientists - opinions soughtUsually, a data scientist has the responsibility of putting models in production, or to interface with a data engineer/software engineer that has that responsibility.
In that case, the JMP Pro ability to generate scoring code in production languages such as SAS, SQL, Python, C, and JavaScript can be very useful.
Re: data scientists - opinions soughtWhile "Data Scientist" isn't as well-defined as, say, "Professional Arm-Wrestler", off the top of my head:
For the "data wrangling" component--
HTTP Request
Query Builder
Python, R, Matlab + SAS integration
Recode + (if Pro) its ability to be deployed to Formula Depot
Table manipulation (stack/split, etc.)
For the non-wrangling components: once the data is in place, I'd say the usual suspects... Graph Builder, Text Explorer, JMP Pro analytic pieces (especially GenReg), and the interactive features that aren't easily accessed in R or Python.
