How can I interpret property based test code?
I'm struggling to translate the instructions on the following snippet:
let myProperty = Prop.forAll fiveAndThrees <| fun number ->
let actual = transform number
let expected = "jbdhjsdhjdsjhsdglsdjlljh"
expected = actual
Check.QuickThrowOnFailure myProperty
Specifically, I'm struggling with the backwards pipeline operator (i.e. "<|").
Here's the test:
[<Fact>]
let ``FizzBuzz.transform returns FizzBuzz`` () =
let fiveAndThrees = Arb.generate<int> |> Gen.map ((*) (3 * 5))
|> Arb.fromGen
let myProperty = Prop.forAll fiveAndThrees <| fun number ->
let actual = transform number
let expected = "jbdhjsdhjdsjhsdglsdjlljh"
expected = actual
Check.QuickThrowOnFailure myProperty
Can someone please guide me step by step on how this code works?
Could this be rewritten using the forward pipe operator (i.e. "|>")?
This answer only covers why <| is used instead of |>.
Here are 5 examples that work toward making use of <| with a large function. The 6th example is to show how the code looks using |> instead of <|. The point is that with the 6th example using |> you have to look into the code to find the primary function funThatNeedsListAndFunc but with <| the primary function funThatNeedsListAndFunc is obvious. So you typically see <| when the last parameter is a function and you want to pass in the function after the primary function for easier comprehension. That's all; don't read more into than that.
After reading the Mark's blog I also learned that <| is useful to remove ( ) around fun. An example using ( ) is given as example 7.
let funThatNeedsListAndFunc list func =
func list
let func = List.sum
let list = Seq.toList { 0 .. 5}
let result1 = funThatNeedsListAndFunc list func
printfn "result1: %A" result1
let result2 = funThatNeedsListAndFunc list <| func
printfn "result2: %A" result2
let result3 = funThatNeedsListAndFunc list <| List.sum
printfn "result3: %A" result3
let result4 = funThatNeedsListAndFunc list <|
fun (list : 'a list) -> List.sum list
printfn "result4: %A" result4
let result5 = funThatNeedsListAndFunc list <|
fun (list : 'a list) ->
// This will be a long function
let a = 1
let b = 2
let c = a * b
let result = List.sum list
let d = "more useless lines"
let (e,f,g) = ("a", 15, 3.0)
result
printfn "result5: %A" result5
.
let result6 =
fun (list : 'a list) ->
// This will be a long function
let a = 1
let b = 2
let c = a * b
let result = List.sum list
let d = "more useless lines"
let (e,f,g) = ("a", 15, 3.0)
result
|> funThatNeedsListAndFunc list
printfn "result6: %A" result6
.
let result7 =
funThatNeedsListAndFunc list (fun (list : 'a list) ->
// This will be a long function
let a = 1
let b = 2
let c = a * b
let result = List.sum list
let d = "more useless lines"
let (e,f,g) = ("a", 15, 3.0)
result)
printfn "result7: %A" result7
Related
How can I get get a list of names visible in the scope with FSC?
I tried this:
#r "../../packages/FSharp.Compiler.Service.16.0.2/lib/net45/FSharp.Compiler.Service.dll"
open Microsoft.FSharp.Compiler
open Microsoft.FSharp.Compiler.SourceCodeServices
do
let file = "TestFileName.fsx"
let checker = SourceCodeServices.FSharpChecker.Create()
let code =
"""
let testStr = "x"
t
"""
async{
let! options, _ = checker.GetProjectOptionsFromScript(file,code)
let! parseRes,checkAnser = checker.ParseAndCheckFileInProject(file, 0, code, options)
match checkAnser with
| FSharpCheckFileAnswer.Succeeded checkRes ->
let! decls =
checkRes.GetDeclarationListInfo(
Some parseRes, //ParsedFileResultsOpt
3 , //line
1 , //colAtEndOfPartialName
"t" , //lineText
[ "t" ] , //qualifyingNames
"" , //partialName
( fun _ -> [] ) //getAllSymbols: (unit -> AssemblySymbol list)
)
if Seq.isEmpty decls.Items then
printfn "*no declarations found*"
else
decls.Items
|> Seq.sortBy (fun d -> d.Name)
|> Seq.truncate 10
|> Seq.iter (fun d -> printfn "decl: %s" d.Name)
| _ -> failwithf "*Parsing did not finish... "
} |> Async.RunSynchronously
but it only prints "no declarations found". I would expect not only testStr but also all the other names that are available by default.
I did not find an example in the documentation.
qualifyingNames should be an empty list, it’s for dot separated prefix, excluding the last (possibly partial) ident. However, there is no a method in FCS that returns unfiltered list of names for scope, yet it’s really easy to add one.
With the answer of vasily-kirichenko and using the current FCS 17.0.1 I came up with this solution:
#r "../../packages/FSharp.Compiler.Service.17.0.1/lib/net45/FSharp.Compiler.Service.dll"
open Microsoft.FSharp.Compiler
open Microsoft.FSharp.Compiler.SourceCodeServices
do
let file = "TestFileName.fsx"
let checker = SourceCodeServices.FSharpChecker.Create()
let code =
"""
let testStr = "x"
testStr.
"""
async{
let! options, _ = checker.GetProjectOptionsFromScript(file,code)
let! parseRes,checkAnser = checker.ParseAndCheckFileInProject(file, 0, code, options)
match checkAnser with
| FSharpCheckFileAnswer.Succeeded checkRes ->
let! decls =
let partialName = PartialLongName.Empty 6 //use any location before before the dot to get all declarations in scope
//let partialName = PartialLongName.Empty 7 //use the loacation of the dot (7) to get memebers of string
checkRes.GetDeclarationListInfo(
Some parseRes, // ParsedFileResultsOpt
3 , // line
"testStr." , // lineText
partialName, // PartialLongName
( fun _ -> [] ) // getAllSymbols: (unit -> AssemblySymbol list)
)
if Seq.isEmpty decls.Items then
printfn "*no declarations found*"
else
decls.Items
|> Seq.sortBy (fun d -> d.Name)
|> Seq.truncate 10
|> Seq.iter (fun d -> printfn "decl: %s" d.Name)
| _ -> failwithf "*Parsing did not finish... "
} |> Async.RunSynchronously
I'm sure this is trivial. However, the following test will always pass:
let transform number =
match number % 3, number % 5 with
| 0, 0 -> "FizzBuzz"
| _, 0 -> "Buzz"
| 0, _ -> "Fizz"
| _ -> number.ToString()
[<Fact>]
let ``FizzBuzz.transform returns FizzBuzz`` () =
let fiveAndThrees = Arb.generate<int> |> Gen.map ((*) (3 * 5))
|> Arb.fromGen
Prop.forAll fiveAndThrees <| fun number ->
let actual = transform number
let expected = "jbdhjsdhjdsjhsdglsdjlljh"
expected = actual
I've also tried:
Check.QuickThrowOnFailure <| (expected = actual)
Specifically, why does this test continue to pass when it should obviously fail?
You're creating the property, but never actually checking it. It just sits there, never executed once.
To check the property, you need to pass it to one of the Check.* methods:
[<Fact>]
let ``FizzBuzz.transform returns FizzBuzz`` () =
let fiveAndThrees = Arb.generate<int> |> Gen.map ((*) (3 * 5))
|> Arb.fromGen
let myProperty = Prop.forAll fiveAndThrees <| fun number ->
let actual = transform number
let expected = "jbdhjsdhjdsjhsdglsdjlljh"
expected = actual
Check.QuickThrowOnFailure myProperty
You can also use FsCheck.Xunit and its [<Property>] attribute:
[<Property(QuietOnSuccess = true)>]
let ``FizzBuzz.transform returns FizzBuzz`` () =
let fiveAndThrees =
Arb.generate<int> |> Gen.map ((*) (3 * 5)) |> Arb.fromGen
Prop.forAll fiveAndThrees <| fun number ->
let actual = FizzBuzz.transform number
let expected = "FizzBuzz"
expected = actual
In the following code, the compiler gets error on index <- index + 1 with error
Error 3 The mutable variable 'index' is used in an invalid way. Mutable variables cannot be captured by closures. Consider eliminating this use of mutation or using a heap-allocated mutable reference cell via 'ref' and '!'. d:\Users....\Program.fs 11 22 ConsoleApplication2
However, it has been defined as mutable?
let rec iterateTupleMemberTypes (tupleArgTypes: System.Type[]) (columnNames: string[]) startingIndex =
seq {
let mutable index = startingIndex
for t in tupleArgTypes do
match t.IsGenericType with
| true -> iterateTupleMemberTypes (t.GetGenericArguments()) columnNames index |> ignore
| false ->
printfn "Name: %s Type: %A" (columnNames.[index]) t
index <- index + 1
yield (columnNames.[index]), t
} |> Map.ofSeq
let myFile = CsvProvider<"""d:\temp\sample.txt""">.GetSample()
let firstRow = myFile.Rows |> Seq.head
let tupleType = firstRow.GetType()
let tupleArgTypes = tupleType.GetGenericArguments()
let m = iterateTupleMemberTypes tupleArgTypes myFile.Headers.Value 0
An idiomatic version of this might look like the following:
#r #"..\packages\FSharp.Data.2.2.2\lib\net40\FSharp.Data.dll"
open FSharp.Data
open System
type SampleCsv = CsvProvider<"Sample.csv">
let sample = SampleCsv.GetSample()
let rec collectLeaves (typeTree : Type) =
seq {
match typeTree.IsGenericType with
| false -> yield typeTree.Name
| true -> yield! typeTree.GetGenericArguments() |> Seq.collect collectLeaves
}
let columnTypes = (sample.Rows |> Seq.head).GetType() |> collectLeaves
let columnDefinitions = columnTypes |> Seq.zip sample.Headers.Value |> Map.ofSeq
let getDefinitions (sample : SampleCsv) = (sample.Rows |> Seq.head).GetType() |> collectLeaves |> Seq.zip sample.Headers.Value |> Map.ofSeq
Personally, I wouldn't be concerned too much about the performance of Map vs Dictionary (and rather have the immutable Map) unless there are hundreds of columns.
The statement after it, let index = 0, shadows your definition of mutable variable index. Also, to make mutables work in sequences, you need refs. https://msdn.microsoft.com/en-us/library/dd233186.aspx
Suggested by #Ming-Tang, I changed the mutable variable to ref and it works now. However, is it a way not to use mutable/ref variable at all?
let rec iterateTupleMemberTypes (tupleArgTypes: System.Type[]) (columnNames: string[]) startingIndex =
seq {
let index = ref startingIndex
for t in tupleArgTypes do
match t.IsGenericType with
| true ->
yield! iterateTupleMemberTypes (t.GetGenericArguments()) columnNames !index
| false ->
printfn "Name: %s Type: %A" (columnNames.[!index]) t
yield (columnNames.[!index]), t
index := !index + 1
} |> dict
let myFile = CsvProvider<"""d:\temp\sample.txt""">.GetSample()
let firstRow = myFile.Rows |> Seq.head
let tupleType = firstRow.GetType()
let tupleArgTypes = tupleType.GetGenericArguments()
let m = iterateTupleMemberTypes tupleArgTypes myFile.Headers.Value 0
I have these types:
type ShouldRetry = ShouldRetry of (RetryCount * LastException -> bool * RetryDelay)
and RetryCount = int
and LastException = exn
and RetryDelay = TimeSpan
type RetryPolicy = RetryPolicy of ShouldRetry
Now I want composability of the retries; something like this:
let serverOverloaded = [| exnRetry<TimeoutException>;
exnRetry<ServerBusyException> |]
|> Array.map (fun fn -> fn (TimeSpan.FromSeconds(4.0)))
let badNetwork = [||] // etc
let compose p1 p2 =
// http://fssnip.net/7h
RetryPolicy(ShouldRetry( (fun (c,e) ->
let RetryPolicy(ShouldRetry(fn)) = p1
let RetryPolicy(ShouldRetry(fn')) = p2
let (cont, delay) = fn c,e
if cont then cont, delay
else
let (cont', delay') = fn' c,e
cont', delay') ))
let finalPolicy = serverOverloaded |> Array.scan compose (RetryPolicies.NoRetry())
But I'm getting compiler errors on fn, delay and fn', saying "The value or constructor 'fn' is not defined".
I can see two problems in your compose function.
When decomposing p1 and p2, the pattern needs to be wrapped in parentheses (otherwise, the compiler interprets the code as a definition of RetryPolicy function, instead of pattern matching):
let (RetryPolicy(ShouldRetry(fn))) = p1
let (RetryPolicy(ShouldRetry(fn'))) = p2
When calling fn' a bit later, you need to pass it the arguments in a tuple (otherwise, the compiler thinks that you're calling fn' with just a single argument c and then building a tuple):
let (cont', delay') = fn' (c,e)
I didn't check (or tried to run) the whole example, so I don't know if the rest of the code does what you want.
First, in order to provide full disclosure, I want to point out that this is related to homework in a Machine Learning class. This question is not the homework question and instead is something I need to figure out in order to complete the bigger problem of creating an ID3 Decision Tree Algorithm.
I need to generate tree similar to the following when given a truth table
let learnedTree = Node(0,"A0", Node(2,"A2", Leaf(0), Leaf(1)), Node(1,"A1", Node(2,"A2", Leaf(0), Leaf(1)), Leaf(0)))
learnedTree is of type BinaryTree which I've defined as follows:
type BinaryTree =
| Leaf of int
| Node of int * string * BinaryTree * BinaryTree
ID3 algorithms take into account various equations to determine where to split the tree, and I've got all that figured out, I'm just having trouble creating the learned tree from my truth table. For example if I have the following table
A1 | A2 | A3 | Class
1 0 0 1
0 1 0 1
0 0 0 0
1 0 1 0
0 0 0 0
1 1 0 1
0 1 1 0
And I decide to split on attribute A1 I would end up with the following:
(A1 = 1) A1 (A1 = 0)
A2 | A3 | Class A2 | A3 | Class
0 0 1 1 0 1
0 1 0 0 0 0
1 0 1 0 0 0
0 1 1
Then I would split the left side and split the right side, and continue the recursive pattern until the leaf nodes are pure and I end up with a tree similar to the following based on the splitting.
let learnedTree = Node(0,"A0", Node(2,"A2", Leaf(0), Leaf(1)), Node(1,"A1", Node(2,"A2", Leaf(0), Leaf(1)), Leaf(0)))
Here is what I've kind of "hacked" together thus far, but I think I might be way off:
let rec createTree (listToSplit : list<list<float>>) index =
let leftSideSplit =
listToSplit |> List.choose (fun x -> if x.Item(index) = 1. then Some(x) else None)
let rightSideSplit =
listToSplit |> List.choose (fun x -> if x.Item(index) = 0. then Some(x) else None)
if leftSideSplit.Length > 0 then
let pureCheck = isListPure leftSideSplit
if pureCheck = 0 then
printfn "%s" "Pure left node class 0"
createTree leftSideSplit (index + 1)
else if pureCheck = 1 then
printfn "%s" "Pure left node class 1"
createTree leftSideSplit (index + 1)
else
printfn "%s - %A" "Recursing Left" leftSideSplit
createTree leftSideSplit (index + 1)
else printfn "%s" "Pure left node class 0"
Should I be using pattern matching instead? Any tips/ideas/help? Thanks a bunch!
Edit: I've since posted an implementation of ID3 on my blog at:
http://blogs.msdn.com/chrsmith
Hey Jim, I've been wanting to write a blog post implementing ID3 in F# for a while - thanks for giving me an execute. While this code doesn't implement the algorithm full (or correctly), it should be sufficient for getting you started.
In general you have the right approach - representing each branch as a discriminated union case is good. And like Brian said, List.partition is definitely a handy function. The trick to making this work correctly is all in determining the optimal attribute/value pair to split on - and to do that you'll need to calculate information gain via entropy, etc.
type Attribute = string
type Value = string
type Record =
{
Weather : string
Temperature : string
PlayTennis : bool
}
override this.ToString() =
sprintf
"{Weather = %s, Temp = %s, PlayTennis = %b}"
this.Weather
this.Temperature
this.PlayTennis
type Decision = Attribute * Value
type DecisionTreeNode =
| Branch of Decision * DecisionTreeNode * DecisionTreeNode
| Leaf of Record list
// ------------------------------------
// Splits a record list into an optimal split and the left / right branches.
// (This is where you use the entropy function to maxamize information gain.)
// Record list -> Decision * Record list * Record list
let bestSplit data =
// Just group by weather, then by temperature
let uniqueWeathers =
List.fold
(fun acc item -> Set.add item.Weather acc)
Set.empty
data
let uniqueTemperatures =
List.fold
(fun acc item -> Set.add item.Temperature acc)
Set.empty
data
if uniqueWeathers.Count = 1 then
let bestSplit = ("Temperature", uniqueTemperatures.MinimumElement)
let left, right =
List.partition
(fun item -> item.Temperature = uniqueTemperatures.MinimumElement)
data
(bestSplit, left, right)
else
let bestSplit = ("Weather", uniqueWeathers.MinimumElement)
let left, right =
List.partition
(fun item -> item.Weather = uniqueWeathers.MinimumElement)
data
(bestSplit, left, right)
let rec determineBranch data =
if List.length data < 4 then
Leaf(data)
else
// Use the entropy function to break the dataset on
// the category / value that best splits the data
let bestDecision, leftBranch, rightBranch = bestSplit data
Branch(
bestDecision,
determineBranch leftBranch,
determineBranch rightBranch)
// ------------------------------------
let rec printID3Result indent branch =
let padding = new System.String(' ', indent)
match branch with
| Leaf(data) ->
data |> List.iter (fun item -> printfn "%s%s" padding <| item.ToString())
| Branch(decision, lhs, rhs) ->
printfn "%sBranch predicate [%A]" padding decision
printfn "%sWhere predicate is true:" padding
printID3Result (indent + 4) lhs
printfn "%sWhere predicate is false:" padding
printID3Result (indent + 4) rhs
// ------------------------------------
let dataset =
[
{ Weather = "windy"; Temperature = "hot"; PlayTennis = false }
{ Weather = "windy"; Temperature = "cool"; PlayTennis = false }
{ Weather = "nice"; Temperature = "cool"; PlayTennis = true }
{ Weather = "nice"; Temperature = "cold"; PlayTennis = true }
{ Weather = "humid"; Temperature = "hot"; PlayTennis = false }
]
printfn "Given input list:"
dataset |> List.iter (printfn "%A")
printfn "ID3 split resulted in:"
let id3Result = determineBranch dataset
printID3Result 0 id3Result
You can use List.partition instead of your two List.choose calls.
http://research.microsoft.com/en-us/um/cambridge/projects/fsharp/manual/FSharp.Core/Microsoft.FSharp.Collections.List.html
(or now http://msdn.microsoft.com/en-us/library/ee353738(VS.100).aspx )
It isn't clear to me that pattern matching will buy you much here; the input type (list of lists) and processing (partitioning and 'pureness' check) doesn't really lend itself to that.
And of course when you finally get the 'end' (a pure list) you need to create a tree, and then presumably this function will create a Leaf when the input only has one 'side' and it's 'pure', but create a Node out of the left-side and right-side results for every other input. Maybe. I didn't quite grok the algorithm completely.
Hopefully that will help steer you a little bit. May be useful to draw up a few smaller sample inputs and outputs to help work out the various cases of the function body.
Thanks Brian & Chris! I was actually able to figure this out and I ended up with the following. This calculates the information gain for determining the best place to split. I'm sure there are probably better ways for me to arrive at this solution especially around the chosen data structures, but this is a start. I plan to refine things later.
#light
open System
let trainList =
[
[1.;0.;0.;1.;];
[0.;1.;0.;1.;];
[0.;0.;0.;0.;];
[1.;0.;1.;0.;];
[0.;0.;0.;0.;];
[1.;1.;0.;1.;];
[0.;1.;1.;0.;];
[1.;0.;0.;1.;];
[0.;0.;0.;0.;];
[1.;0.;0.;1.;];
]
type BinaryTree =
| Leaf of int
| Node of int * string * BinaryTree * BinaryTree
let entropyList nums =
let sumOfnums =
nums
|> Seq.sum
nums
|> Seq.map (fun x -> if x=0.00 then x else (-((x/sumOfnums) * Math.Log(x/sumOfnums, 2.))))
|> Seq.sum
let entropyBinaryList (dataListOfLists:list<list<float>>) =
let classList =
dataListOfLists
|> List.map (fun x -> x.Item(x.Length - 1))
let ListOfNo =
classList
|> List.choose (fun x -> if x = 0. then Some(x) else None)
let ListOfYes =
classList
|> List.choose (fun x -> if x = 1. then Some(x) else None)
let numberOfYes : float = float ListOfYes.Length
let numberOfNo : float = float ListOfNo.Length
let ListOfNumYesAndSumNo = [numberOfYes; numberOfNo]
entropyList ListOfNumYesAndSumNo
let conditionalEntropy (dataListOfLists:list<list<float>>) attributeNumber =
let NoAttributeList =
dataListOfLists
|> List.choose (fun x -> if x.Item(attributeNumber) = 0. then Some(x) else None)
let YesAttributeList =
dataListOfLists
|> List.choose (fun x -> if x.Item(attributeNumber) = 1. then Some(x) else None)
let numberOfYes : float = float YesAttributeList.Length
let numberOfNo : float = float NoAttributeList.Length
let noConditionalEntropy = (entropyBinaryList NoAttributeList) * (numberOfNo/(numberOfNo + numberOfYes))
let yesConditionalEntropy = (entropyBinaryList YesAttributeList) * (numberOfYes/(numberOfNo + numberOfYes))
[noConditionalEntropy; yesConditionalEntropy]
let findBestSplitIndex(listOfInstances : list<list<float>>) =
let IGList =
[0..(listOfInstances.Item(0).Length - 2)]
|> List.mapi (fun i x -> (i, (entropyBinaryList listOfInstances) - (List.sum (conditionalEntropy listOfInstances x))))
IGList
|> List.maxBy snd
|> fst
let isListPure (listToCheck : list<list<float>>) =
let splitList = listToCheck |> List.choose (fun x -> if x.Item(x.Length - 1) = 1. then Some(x) else None)
if splitList.Length = listToCheck.Length then 1
else if splitList.Length = 0 then 0
else -1
let rec createTree (listToSplit : list<list<float>>) =
let pureCheck = isListPure listToSplit
if pureCheck = 0 then
printfn "%s" "Pure - Leaf(0)"
else if pureCheck = 1 then
printfn "%s" "Pure - Leaf(1)"
else
printfn "%A - is not pure" listToSplit
if listToSplit.Length > 1 then // There are attributes we can split on
// Chose best place to split list
let splitIndex = findBestSplitIndex(listToSplit)
printfn "spliting at index %A" splitIndex
let leftSideSplit =
listToSplit |> List.choose (fun x -> if x.Item(splitIndex) = 1. then Some(x) else None)
let rightSideSplit =
listToSplit |> List.choose (fun x -> if x.Item(splitIndex) = 0. then Some(x) else None)
createTree leftSideSplit
createTree rightSideSplit
else
printfn "%s" "Not Pure, but can't split choose based on heuristics - Leaf(0 or 1)"