Taking a Hiatus From Blogging

I have blogged every Tuesday for the last six years.  Thank you to all of the people who took the time to read my posts and give comments.  I hope you benefitted from the content (if not the rookie formatting and sometimes creative spelling/grammar).  Recently, I signed a deal with  a publisher to write a book that has a a pretty aggressive deadline.  Because of the time commitment that writing a book takes,  I am taking a hiatus from blogging.  If all goes to plan, I will resume blogging in mid-2016.


Using Mocking Frameworks To Help With Unit Testing UI Controls

One of the more common reasons that developers tell us of why they don’t unit test is “All of my application is visual controls with code behind. Refactoring all of that code to a .dll that can be united tested will take more time than it is worth.” While it is true that unit testing is easier if your application lives “in code” as a separate assembly, you can still use unit testing in a UI-heavy code base. By judiciously using a mocking framework, you can speed up the process even more.

Consider this form from a WinForm application written in VB.NET.


The grid view has the following code behind :

Private Sub Button1_Click(sender As Object, e As EventArgs) Handles Button1.Click For counter = 0 To DataGridView1.RowCount - 1 If (DataGridView1.Rows(counter).Cells(11).FormattedValue) Then If (DataGridView1.Rows(counter).Cells(10).Value <> "") Then TextBox1.Text = FormatCurrency(TextBox1.Text - DataGridView1.Rows(counter).Cells(10).Value, 2) End If End If Next End Sub


The business logic is intermingled with the visual controls (TextBox1, DataGridView1, etc…). Is there a way to easily unit test this code? The answer is yes. Step one is to add a unit test project to the solution. Step two is to break the sub method into a function method. Once the methods have an input and an output, you can put a unit test on it. For example:

Private Sub Button1_Click(sender As Object, e As EventArgs) Handles Button1.Click For counter = 0 To DataGridView1.RowCount - 1 Dim initialValue = TextBox1.Text Dim cell0 = DataGridView1.Rows(counter).Cells(11) Dim cell1 = DataGridView1.Rows(counter).Cells(10) TextBox1.Text = GetCalculatedValue(initialValue, cell0, cell1) Next End Sub Public Function GetCalculatedValue(initalValue As String, cell0 As DataGridViewCell, cell1 As DataGridViewCell) As String Dim returnValue As String = initalValue If (cell0.FormattedValue) Then If (cell1.Value <> "") Then initalValue = FormatCurrency(TextBox1.Text - cell1.Value, 2) End If End If Return initalValue End Function

And we can add a unit test like this:


[TestMethod] public void GetCalculatedValue_ReturnsExpected() { Form1 instance = new Form1(); String baseValue = "$10.00"; DataGridView gridView = new DataGridView(); gridView.Columns.Add("TEST1", "TEST1"); gridView.Columns.Add("TEST2", "TEST2"); gridView.Rows.Add(new DataGridViewRow()); gridView.Rows[0].Cells[0].Value = "$1.00"; gridView.Rows[0].Cells[1].Value = "$2.00"; DataGridViewCell cell0 = gridView.Rows[0].Cells[0]; DataGridViewCell cell1 = gridView.Rows[0].Cells[1]; var actual = instance.GetCalculatedValue(baseValue, cell0, cell1); var expected = "$8.00"; Assert.AreEqual(expected, actual); }


Although this test runs green, it is suboptimal because we have to standup lots of objects (DataGridView, Columns, DataGridRow) just to get to the class we are interested in, in this case DataGridViewCell. Instead of generating all of that superfluous code, there is a better way to set the state of only the class we want – enter Mocking frameworks. Mocking frameworks give us the ability to focus only on the subjects under test (SUT) while ignoring everything else.

But there is a catch. There are 2 types of mocking frameworks: ones that generate their code based on inspecting the types and ones that generate their code based on the compiled IL. The former group includes RhinoMocks and Moq . If you try and add Moq to this unit test project and generate a DataGridViewCell like this:

[TestMethod] public void GetCalculatedValue_ReturnsExpected() { Form1 instance = new Form1(); String baseValue = "$10.00"; var mock0 = new Mock<DataGridViewCell>(); mock0.SetupGet(dataGridViewCell => dataGridViewCell.Value).Returns("$1.00"); var mock1 = new Mock<DataGridViewCell>(); mock1.SetupGet(dataGridViewCell => dataGridViewCell.Value).Returns("$2.00"); var actual = instance.GetCalculatedValue(baseValue, mock0.Object, mock1.Object); var expected = "$8.00"; Assert.AreEqual(expected, actual); }

You will get an exception



Since we don’t control DataGridViewCell’s code, there is no way to change those properties to overidable/nonvirtual. As a general rule, you only use RhinoMocks/Moq on classes that you can control.

The other type of mocking framework (based on IL) can solve this problem. There are 2 commercial frameworks (JustMock, TypeMock) but they cost $399/year (as of this writing). There is a 3rd framework we can use and it is built into Visual Studio 2012+. It is called the Microsoft Fakes Framework. By adding this to your test project,


you can craft your unit test like so:

[TestMethod] public void GetCalculatedValue_ReturnsExpected() { Form1 instance = new Form1(); String baseValue = "$10.00"; using (ShimsContext.Create()) { var cell0 = new ShimDataGridViewCell(new StubDataGridViewCell()); cell0.FormattedValueGet = () => { return "$1.00"; }; var cell1 = new ShimDataGridViewCell(new StubDataGridViewCell()); cell1.ValueGet = () => { return "$2.00"; }; var actual = instance.GetCalculatedValue(baseValue, cell0, cell1); var expected = "$8.00"; Assert.AreEqual(expected, actual); } }

and get green. The downside of using Microsoft Fakes is that you need to re-generate the fakes if the code changes. This makes it ideal for faking external libraries that don’t change much (like ADO.NET) but not assemblies that are under active development.

EjectABed Version 2 – Now Using the Raspberry Pi (Part 2)

With the connection from Twitter to the PI working well, I decided to hook up the bed top the PI.  The Bed is controlled via a server attached to a bellow that forces air to the screw drive.  You can read about how we figured that one out here.
My initial thought was that it would be easy as the Netduino implementation to control the servo was all of 5 lines of code.  The Netduino has built-in PWM ports and the api has a PWM class:
1 uint period = 20000; 2 uint duration = SERVO_NEUTRAL; 3 _servo = new PWM(PWMChannels.PWM_PIN_D5, period, duration, PWM.ScaleFactor.Microseconds, false); 4 _servo.Start(); 5 _servoReady = true;

However, when I went to look for a PWM port, there wasn’t one!  Ugh!  I want over to Stack Overflow to confirm with this question and sure enough, no PWM.  The only example for servo control that the Windows 10 code samples have are using the GPIO to activate a servo forwards and backwards, but that will not work because I need to hold the bellow in a specific place for the air to push correctly.  The Windows IoT team suggested that I use the AdaFruit PWM shield for the control


So I ordered a couple and then my son soldered the pins in

20150904_201350 20150904_201104

I then hooked up the shield to the servo and the PI
and went to look for some PI code to control the pwms.  Another problem, there isn’t any!  I went over to the Raspberry Pi forums and it turns out, they are waiting for MSFT to finish that piece.  Ugh, I decided to take the path of least resistance and I removed that PWM shield and added back in the Netduino


Now I have the ability to control the servo from the PI.  I would have rather cut out the Netduino completely, but the limitations of Win10 on Raspberry Pi won’t allow me to do that.  Oh well, it is still a good entry and it was a lot of fun to work on.

EjectABed Version 2 – Now Using the Raspberry Pi (Part 1)

I recently entered a hackster.io competition that centered around using Windows 10 on the Raspberry Pi.  I entered the ejectabed and it was accepted to the semi-final round.  My thought was to take the existing ejectabed controller from a Netduino and move it to a Raspberry Pi.  While doing that, I could open the ejectabed from my local area network to the internet so anyone could eject Sloan.
My 1st step was hook my Raspberry Pi up to my home network and deploy from Visual Studio to it.  Turns out, it was pretty straightforward.
I took a old Asus Portable Wireless Router and plugged it into my home workstation.  I then configured the router to act as an Access Point so that it would pass though all traffic from the router to which my developer workstation is attached.  I then attached the router to the PI and powered it though the PI’s USB port.  I then plugged the PI’s HDMI out to a spare monitor of mine.


With all of the hardware plugged in, I headed over to Windows On Devices and followed the instructions on how to set up a Raspberry PI.  After installing the correct software on my developer workstation, flashing the SD card with win10, plugging the SD card into the PI, turning the PI on, and then remoting into the PI via powershell, I could see the PI on my local workstation via the Windows IoT Core Watcher and the PI showing its friendly welcome screen via HDMI.



I then headed over to Visual Studio and copy/pasted the equisite “Hello IoT World” Blinky project to the Pi and watched the light go on and off.


With that out of the way, I decided to look at controlling the light via Twitter and Azure.  The thought was to have the PI monitor a message queue on Azure and whenever there was a message, blink on or off (simulating the ejectabed being activated).  To that end, I went into Azure and created a basic storage account.  One of the nice things about Azure is that you get a queue out of the box when you create a storage account:


One of the not so nice things about Azure is that there is no way to control said Queue via their UI.  You have to create, push, and pull from the queue in code.  I went back to visual studio and added in the Azure Storage Nuget package


I then created a method to monitor the queue
1 internal async Task<Boolean> IsMessageOnQueue() 2 { 3 var storageConnectionString = "DefaultEndpointsProtocol=https;AccountName=ejectabed;AccountKey=xxx"; 4 var storageAccount = CloudStorageAccount.Parse(storageConnectionString); 5 var client = storageAccount.CreateCloudQueueClient(); 6 var queue = client.GetQueueReference("sloan"); 7 var queueExists = await queue.ExistsAsync(); 8 if (!queueExists) 9 { 10 GpioStatus.Text = "Queue does not exist or is unreachable."; 11 return false; 12 } 13 var message = await queue.GetMessageAsync(); 14 if (message != null) 15 { 16 await queue.DeleteMessageAsync(message); 17 return true; 18 } 19 GpioStatus.Text = "No message for the EjectABed."; 20 return false; 21 } 22

Then if there is a message, the PI would run the ejection sequence (in this case blink the light)
1 internal void RunEjectionSequence() 2 { 3 bedCommand.Eject(); 4 bedTimer = new DispatcherTimer(); 5 bedTimer.Interval = TimeSpan.FromSeconds(ejectionLength); 6 bedTimer.Tick += LightTimer_Tick; 7 bedTimer.Start(); 8 }


I deployed the code to the PI without a problem.  I then created a Basic console application to push messages to the queue that the PI could drain
1 class Program 2 { 3 static String storageConnectionString = "DefaultEndpointsProtocol=https;AccountName=ejectabed;AccountKey=xxx"; 4 5 static void Main(string[] args) 6 { 7 Console.WriteLine("Start"); 8 Console.WriteLine("Press The 'E' Key To Eject. Press 'Q' to quit..."); 9 10 var keyInfo = ConsoleKey.S; 11 do 12 { 13 keyInfo = Console.ReadKey().Key; 14 if (keyInfo == ConsoleKey.E) 15 { 16 CreateQueue(); 17 WriteToQueue(); 18 //ReadFromQueue(); 19 } 20 21 } while (keyInfo != ConsoleKey.Q); 22 23 Console.WriteLine("End"); 24 Console.ReadKey(); 25 } 26 27 private static void CreateQueue() 28 { 29 var storageAccount = CloudStorageAccount.Parse(storageConnectionString); 30 var client = storageAccount.CreateCloudQueueClient(); 31 var queue = client.GetQueueReference("sloan"); 32 queue.CreateIfNotExists(); 33 Console.WriteLine("Created Queue"); 34 } 35 36 private static void WriteToQueue() 37 { 38 var storageAccount = CloudStorageAccount.Parse(storageConnectionString); 39 var client = storageAccount.CreateCloudQueueClient(); 40 var queue = client.GetQueueReference("sloan"); 41 var message = new CloudQueueMessage("Eject!"); 42 queue.AddMessage(message); 43 Console.WriteLine("Wrote To Queue"); 44 } 45 46 47 private static void ReadFromQueue() 48 { 49 var storageAccount = CloudStorageAccount.Parse(storageConnectionString); 50 var client = storageAccount.CreateCloudQueueClient(); 51 var queue = client.GetQueueReference("sloan"); 52 var queueExists = queue.Exists(); 53 if (!queueExists) 54 Console.WriteLine("Queue does not exist"); 55 var message = queue.GetMessage(); 56 if (message != null) 57 { 58 queue.DeleteMessage(message); 59 Console.WriteLine("Message Found and Deleted"); 60 } 61 else 62 { 63 Console.WriteLine("No messages"); 64 } 65 } 66

I could then Write to the queue and the PI would read and react.  You can see it in action here:


With the queue up and running, I was ready to add in the ability for someone to Tweet to the queue.  I created a cloud service project and pointed to a new project that will monitor Twitter and then push to the queue:



The Twitter project uses the TweetInvi nuget package and is a worker project.  It makes a call to Twitter every 15 seconds and if there is a tweet to “ejectabed” with a person’s name, it will write to the queue (right now, only Sloan’s name is available)
1 type TwitterWorker() = 2 inherit RoleEntryPoint() 3 4 let storageConnectionString = RoleEnvironment.GetConfigurationSettingValue("storageConnectionString") 5 6 let createQueue(queueName) = 7 let storageAccount = CloudStorageAccount.Parse(storageConnectionString) 8 let client = storageAccount.CreateCloudQueueClient() 9 let queue = client.GetQueueReference(queueName); 10 queue.CreateIfNotExists() |> ignore 11 12 let writeToQueue(queueName) = 13 let storageAccount = CloudStorageAccount.Parse(storageConnectionString) 14 let client = storageAccount.CreateCloudQueueClient() 15 let queue = client.GetQueueReference(queueName) 16 let message = new CloudQueueMessage("Eject!") 17 queue.AddMessage(message) |> ignore 18 19 let writeTweetToQueue(queueName) = 20 createQueue(queueName) 21 writeToQueue(queueName) 22 23 let getKeywordFromTweet(tweet: ITweet) = 24 let keyword = "sloan" 25 let hasKeyword = tweet.Text.Contains(keyword) 26 let isFavourited = tweet.FavouriteCount > 0 27 match hasKeyword, isFavourited with 28 | true,false -> Some (keyword,tweet) 29 | _,_ -> None 30 31 32 override this.Run() = 33 while(true) do 34 let consumerKey = RoleEnvironment.GetConfigurationSettingValue("consumerKey") 35 let consumerSecret = RoleEnvironment.GetConfigurationSettingValue("consumerSecret") 36 let accessToken = RoleEnvironment.GetConfigurationSettingValue("accessToken") 37 let accessTokenSecret = RoleEnvironment.GetConfigurationSettingValue("accessTokenSecret") 38 39 let creds = Credentials.TwitterCredentials(consumerKey, consumerSecret, accessToken, accessTokenSecret) 40 Tweetinvi.Auth.SetCredentials(creds) 41 let matchingTweets = Tweetinvi.Search.SearchTweets("@ejectabed") 42 let matchingTweets' = matchingTweets |> Seq.map(fun t -> getKeywordFromTweet(t)) 43 |> Seq.filter(fun t -> t.IsSome) 44 |> Seq.map (fun t -> t.Value) 45 matchingTweets' |> Seq.iter(fun (k,t) -> writeTweetToQueue(k)) 46 matchingTweets' |> Seq.iter(fun (k,t) -> t.Favourite()) 47 48 Thread.Sleep(15000) 49 50 override this.OnStart() = 51 ServicePointManager.DefaultConnectionLimit <- 12 52 base.OnStart()

Deploying to Azure was a snap
And now when I Tweet,
the PI reacts.  Since Twitter does not allow the same Tweet to be sent again, I deleted it every time I wanted to send a new message to the queue.

Facebook Api Using F#

A common requirement for modern user-facing applications is to interface with Facebook.  Unfortunately, Facebook does not make it easy on developers –> in fact it is one of the harder apis that I have seen.  However, there is a covering sdk that you can use, along with some hoop jumping, to get it working.  The problem is one of assumptions.  The .NET sdk assumes that you want to build a Windows Store or Phone app and it is human to facebook connections.  Once you get past those assumptions, you can do pretty well.

The first thing you need to do is set up a Facebook account.



Then register as a developer and create an application


In Visual Studio, Nuget in the facebook sdk


Then, in the REPL add the following code to get the auth token

1 #r "../packages/Facebook.7.0.6/lib/net45/Facebook.dll" 2 #r "../packages/Newtonsoft.Json.7.0.1/lib/net45/Newtonsoft.Json.dll" 3 4 open Facebook 5 open Newtonsoft.Json 6 7 type Credentials = {client_id:string; client_secret:string; grant_type:string;scope:string} 8 let credentials = {client_id="123456"; 9 client_secret="123456"; 10 grant_type="client_credentials"; 11 scope="manage_pages,publish_stream,read_stream,publish_checkins,offline_access"} 12 13 14 let client = FacebookClient() 15 let tokenJson = client.Get("oauth/access_token",credentials) 16 type Token = {access_token:string} 17 let token = JsonConvert.DeserializeObject<Token>(tokenJson.ToString());

Which gives


Once you get the token, you can make a request to user and post to the page

1 let client' = FacebookClient(token.access_token) 2 client'.Get("me") 3 4 let pageId = "me" 5 type FacecbookPost = {title:string; message:string} 6 let post = {title="Test Title"; message = "Test Message"} 7 client'.Post(pageId + "/feed", post) 8

I was getting this message though


So then the fun part.  Apparently, you need to submit your application to the facebook team to be approved to be used.  So now I have to submit icons and a description on how this application will be used before I can make a POST.  <sigh>

Thanks to Gene Belitski for his help on my question on Stack Overflow

Wake County Voter Analysis Using FSharp, AzureML, and R

One of the real strengths of FSharp its ability to plow through and transform data in a very intuitive way,  I was recently looking at Wake Country Voter Data found here to do some basic voter analysis.  My first thought was to download the data into R Studio.  Easy?  Not really.  The data is available as a ginormous Excel spreadsheet of database of about 154 MB in size.  I wanted to slim the dataset down and make it a .csv for easy import into R but using Excel to export the data as a .csv kept screwing up the formatting and importing it directly into R Studio from Excel resulting in out of memory crashes.  Also, the results of the different election dates were not consistent –> sometimes null, sometimes not.   I managed to get the data into R Studio without a crash and wrote a function of either voted “1” or not “0” for each election

1 #V = voted in-person on Election Day 2 #A = voted absentee by mail or early voting (through May 2006) 3 #M = voted absentee by mail (November 2006 - present) 4 5 #O = voted One-Stop early voting (November 2006 - present) 6 #T = voted at a transfer precinct on Election Day 7 #P = voted a provisional ballot 8 #L = Legacy data (prior to 2006) 9 #D = Did not show 10 11 votedIndicated <- function(votedCode) { 12 switch(votedCode, 13 "V" = 1, 14 "A" = 1, 15 "M" = 1, 16 "O" = 1, 17 "T" = 1, 18 "P" = 1, 19 "L" = 1, 20 "D" = 0) 21 } 22

However, every time I tried to run it, the IDE would crash with an out of memory issue. 

 Stepping back, I decided to transform the data in Visual Studio using FSharp. I created a sample from the ginormous excel spreadsheet and then imported the data using a type provider.  No memory crashes!

1 #r "../packages/ExcelProvider.0.1.2/lib/net40/ExcelProvider.dll" 2 open FSharp.ExcelProvider 3 4 [<Literal>] 5 let samplePath = "../../Data/vrdb-Sample.xlsx" 6 7 open System.IO 8 let baseDirectory = __SOURCE_DIRECTORY__ 9 let baseDirectory' = Directory.GetParent(baseDirectory) 10 let baseDirectory'' = Directory.GetParent(baseDirectory'.FullName) 11 let inputFilePath = @"Data\vrdb.xlsx" 12 let fullInputPath = Path.Combine(baseDirectory''.FullName, inputFilePath) 13 14 type WakeCountyVoterContext = ExcelFile<samplePath> 15 let context = new WakeCountyVoterContext(fullInputPath) 16 let row = context.Data |> Seq.head

I then applied a similar function for voted or not and then exported the data as a .csv

1 let voted (voteCode:obj) = 2 match voteCode = null with 3 | true -> "0" 4 | false -> "1" 5 6 open System 7 let header = "Id,Race,Party,Gender,Age,20080506,20080624,20081104,20091006,20091103,20100504,20100622,20101102,20111011,20111108,20120508,20120717,20121106,20130312,20131008,20131105,20140506,20140715,20141104" 8 9 let createOutputRow (row:WakeCountyVoterContext.Row) = 10 String.Format("{0},{1},{2},{3},{4},{5},{6},{7},{8},{9},{10},{11},{12},{13},{14},{15},{16},{17},{18},{19},{20},{21},{22},{23}", 11 row.voter_reg_num, 12 row.race_lbl, 13 row.party_lbl, 14 row.gender_lbl, 15 row.eoy_age, 16 voted(row.``05/06/2008``), 17 voted(row.``06/24/2008``), 18 voted(row.``11/04/2008``), 19 voted(row.``10/06/2009``), 20 voted(row.``11/03/2009``), 21 voted(row.``05/04/2010``), 22 voted(row.``06/22/2010``), 23 voted(row.``11/02/2010``), 24 voted(row.``10/11/2011``), 25 voted(row.``11/08/2011``), 26 voted(row.``05/08/2012``), 27 voted(row.``07/17/2012``), 28 voted(row.``11/06/2012``), 29 voted(row.``03/12/2013``), 30 voted(row.``10/08/2013``), 31 voted(row.``11/05/2013``), 32 voted(row.``05/06/2014``), 33 voted(row.``07/15/2014``), 34 voted(row.``11/04/2014``) 35 ) 36 37 let outputFilePath = @"Data\vrdb.csv" 38 39 let data = context.Data |> Seq.map(fun row -> createOutputRow(row)) 40 let fullOutputPath = Path.Combine(baseDirectory''.FullName, outputFilePath) 41 42 let file = new StreamWriter(fullOutputPath,true) 43 44 file.WriteLine(header) 45 context.Data |> Seq.map(fun row -> createOutputRow(row)) 46 |> Seq.iter(fun r -> file.WriteLine(r)) 47

The really great thing is that I could write and then dispose of each line so I could do it without any crashes.  Once the data was into a a .csv (10% the size of Excel), I could then import it into R Studio without a problem.  It is a common lesson but really shows that using the right tool for the job saves tons of headaches.

I knew from a previous analysis of voter data that the #1 determinate of a person from wake county voting in a off-cycle election was their age:




So then in R, I created a decision tree for just age to see what the split was:

1 library(rpart) 2 temp <- rpart(all.voters$X20131008 ~ all.voters$Age) 3 plot(temp) 4 text(temp)

Thanks to Placidia for answering my question on stats.stackoverflow


So basically politicians should be targeting people 50 years or older or perhaps emphasizing issues that appeal to the over 50 crowd.





Kaggle and R

Following up on last week’s post on doing a Kaggle competition, I then decided to see if I could explore the data more in R on my local desktop.  The competition is about analyzing a large group of house claims to give them a risk score.

I started the R studio to take a look at the initial data:

1 train <- read.csv("../Data/train.csv") 2 head(train) 3 summary(train) 4 5 plot(train$Hazard)


A couple of things popped out.  All of the X variables look to be categorical.  Even the result “Hazard” is an integer with most of the values falling between 1 and 9.

With that in mind, I decided to split the dataset into two sections: the majority and the minority.

1 train.low <- subset(train, Hazard < 9) 2 train.high <- subset(train, Hazard >= 9) 3 4 plot(train.low$Hazard) 5 plot(train.high$Hazard)

With the under as:


And the over 9 is like this


But I want to look at the Hazard score from a distribution point of view:

1 hazard.frame <- as.data.frame(table(train$Hazard)) 2 colnames(hazard.frame) <- c("hazard","freq") 3 hist(hazard.frame$freq) 4 plot(x=hazard.frame$hazard, y=hazard.frame$freq) 5 plot(x=hazard.frame$hazard, log(y=hazard.frame$freq)) 6

The hist shows the left skew




and the log plot really shows the distribution


So there is clearly a diminishing return going on.   As of this writing, the leader is at 40%, which is about 20,400 of the 51,000 entries.   So if you could identify all of the ones correctly, you should get 37% of the way there.  To test it out, I submitted to Kaggle only ones:


LOL, so they must take away for incorrect answers as it is same as “all 0” benchmark.  So going back, I know that if I can predict the ones correctly and make a reasonable guess at the rest, I might be OK.   I went back and tuned my model some to get me out of the bottom 25% and then let it be.  I assume that there is something obvious/industry standard that I am missing because there are so many people between my position and the top 25%.


Get every new post delivered to your Inbox.

Join 26 other followers