Now, AI companies are obliged to tell you how great their model is. They may say something like “our model is 95% accurate”. Zowee! But what does this mean exactly? In terms of binary classification it means that the model chose the correct class 95% of the time. This seems pretty good, so what’s the problem?
Suppose I create an AI that guesses the gender of a technical employee at Facbook. As of 2017, 19% of STEM roles are held by women. Behind the scenes, my model is really simple: it just chooses male every time (bonus question: is this AI?). Because of the data, my model will be 81% accurate. Now 95% doesn’t seem all that impressive. This dataset is known to be unbalanced, because the classes are not proportional. A better dataset would have about 50% women and 50% men. So asking if a dataset is balanced helps to identify some tricks that make models appear smarter than they are.
With wildly unbalanced data (like diagnosing rare diseases), measures like positive predictive value are far more important than overall accuracy.
When I need to quickly create smaller sets of test data or dummy data, I use Mockaroo. It is highly configurable with over 140 built-in field types for locations, personal information, product information, technical information and much more. Every field type can be customized, and you can also use your own regular expression to generate data. The data can then be exported to CSV, JSON, SQL, and Excel formats. The interface is simple to use and understand, and you can save your schemas and data sets for later reuse.
I’m fond of Coblis and was aware of the last two, but the first two were new to me.
In my previous post, I have demonstrated how easy it is to create a bar graph in SQL Server 2017 In-Database Machine Learning using R.
We’re going to build upon that basic graph.
Sometimes doing data analysis would require us to look at an overview of our data across specific partitions, say a year. For example, we want to see how our product groups fare on month-to-month basis across the last 4 years.
In a data analytics perspective, there are quite a handful of data points in this requirement – data aggregate (quantity), monthly periods, and year partitions.
One of the approaches to handle such requirement is by using a
facet. Faceting is a way of plotting subsets of data into a matrix of panels based on one or more variables – or facets.
Click through for the example and code. Facets are quite useful, but they run the risk of misleading if you squeeze too many onto the screen. The same line can look quite different with a “tall” facet versus a “wide” facet, and that can change how people interpret your visual.
The Admin stack is probably the most important stack here. You still using maintenance tasks via SSMS? stop doing that. Rebuilding indexes every night? Maybe rethink that.
How you keep track of, monitor and do basics DBA tasks?
Ok so this can involve SSMS, but a feature not a lot of people may not use. We use it to keep track of all of our instances and push things..oh baby baaaby! It also allows me to combine PoSh to do work against instances, gather data (historical, dmv…etc) and do a boat load of admin stuff without pointing and clicking. Heck I don’t even have to open SSMS to use my CMS server at all.
SQLSentry can automatically defrag indexes for you and update stats. You could use this instead of the below choices for this aspect if desired. Although not free, it’s an option we have in our environment and I love me some options.
Both amazing options for backup, reindexing and checkdb. Although most places i’ve worked use Ola’s scripts by default. HOWEVER…. Minion has some pretty nice options that are FAR more configurable than Ola’s. We have mitigated some large DB issues by rolling our own code on top of Ola’s scripts. We could avoid this by simply using Minionware!
I’m a huge fan of the Minionware suite. And several other things Warren mentions.
Recently I was doing some work with a friend around some database restores. It was pretty routine stuff. However, after one restore my friend came across something in the SQL Error Log that caught him by surprise. As part of the restore, there was a CHECKDB message for the restored database:
My friend’s first reaction was “why is SQL Server doing a DBCC CHECKDB as part of the restore?” He was concerned, because CHECKDB is a pretty hefty operation and this could really impact the restore time if he had to wait on a CHECKDB to complete. But the other confusing thing was that the date for the CHECKDB didn’t match up with the restore timing.
Click through to learn the answer.
Documentation is an important part of every DBA’s job and Snagit Screen Capture is my screenshot tool of choice. I don’t know how I ever lived without it. Snagit is awesome for both pics and videos – wayyy better than the built-in (but still useful) Snipping tool.
Click through for several other tooling recommendations.
This T-SQL Tuesday is brought to us by Jens Vestergaard (b | t), and we are asked to share our favorite SQL Server tools. Hint Profiler will not be on the list. But where do you start there are so many tools out there. In alphabetical order here are my top 5 tools because I can’t pick which one is better than other.
Click through to see Tracy’s top 5 list.
So, I bit: the tweet he referenced was announcing a new version of the SQL Server module (21.0.17240). Here’s a quick list of the updates included:
- Added Get-SqlBackupHistory cmdlet
- Ported PS Provider to .NET Core for PowerShell 6 support
- Ported a subset of cmdlets to .NET Core for PowerShell 6 support
- Powershell 6 support on macOS and Linux in Preview.
- To use SqlServer provider on macOS and Linux mount it using a new PSDrive. Examples in documentation.
- Removed restriction of 64-bit OS for this module. Note: Invoke-Sqlcmd cmdlet is the only cmdlet not supported on 32-bit OS.
The bold lines are my emphasis: with PowerShell 6 support for Linux and macOS, that opens up new avenues for connecting to and automating SQL Server from any platform. This is exciting stuff. I couldn’t wait to take it for a spin, so I set up a quick demo environment to test it out.
It’s not perfect but it did give Drew the vapors, which is a good sign that they’re on the right track.
Using the default trace is pretty lightweight and backwards compatible. While I generally try to avoid traces, I like this method because it doesn’t require remote access, it works on older SQL instances, it’s accurate and reading from the trace isn’t as CPU-intensive as it would be with an Extended Event.
Click through for details on this as well as three other methods, along with the dbatools glue to make it work.