Loading collection data...
Collections are a way for you to organize kata so that you can create your own training routines. Every collection you create is public and automatically sharable with other warriors. After you have added a few kata to a collection you and others can train on the kata contained within the collection.
Get started now by creating a new collection.
This comment is hidden because it contains spoiler information about the solution
Neat indeed!
Wow, that is elegant!
'alr approved some time ago'
I still don't get where the
(b - 1) * fib(n + 2) = (b - 1) * fib(n) + (b - 1) * fib(n + 1)
comes from and why should I add this to the original equation. Please, explain in details or give me link to a book or some article where it is well-explained.@Technetium_Phenol Oh, C'mon :D It's 21st century, forget Fortran and come to Scala :D
The description is not correct.
If one translates the above definition directly into a recursive function, the result is not very efficient. One can try memoization, but that requires lots of space and is not necessary. So, first step is to try and find a tail recursive definition. In order to do that we try to write both sides of equation 3) on the same form. Currently, the left side of the equation contains a single term, whereas the right side is the sum of two terms. A first attempt is to add fib(n + 1) to both sides of the equation:they have fib(0) = 1
and there code doesn't calculate the fibonnaci series.
Here is my alternative expansion.
The function is recursively defined by:
3. fib(n + 1) + fib(n + 2) = fib(n) + 2 * fib(n + 1)
The two sides of the equation look much more alike, but there is still an essential difference, which is the coefficient of the second term of each side. On the left side of the equation, it is 1 and, on the right, it is 2. To remedy this, we can introduce a variable b:
3. b * fib(n + 1) + b * fib(n + 2) = b * fib(n) + (b + b) * fib(n + 1)
We then subtract both sides by b * fib(n+1)
3. b * fib(n + 2) = b * fib(n) + (b + b)
To try and return the tail to the function we add both sides by a * fib(n+1) to each side:
3. a * fib(n + 1) + b * fib(n + 2) = b * fib(n) + (a + b) * fib(n + 1)
Now the two sides have the same form (call it F), this function has the form:
3. F(c, d, e) = F(a, b, n+1) = F(a, a+b, n) = c * fib(e) + d * fib(e + 1)
(Note if your confused by the above equation substitute the values into the furthest right function and you get)
3. a * fib(n + 1) + b * fib(n + 2) = b * fib(n) + (a + b) * fib(n + 1) = c * fib(n) + d * fib(n + 1)
Which we can simplify to:
3. F(a, b, n + 1) = F(b, a + b, n)
We also have, by definition of F and fib:
4. F(a, b, 0) = a * fib(0) + b * fib(1) = b
Also, by definition of F:5\. fib(n) = F(1, 0, n) with a = 1, b = 0
6\. F(a, b, 1) = a * fib(1) + b * fib(2) = a + b
The next step is to translate the above into code:
The final step (optional for languages that support tail call optimization) is to replace the tail recursive function F with a loop:
(1) that each golfer plays "exactly once every day"
That means same set of players everyday.
GO Language Translation Added :D
I made the same mistake, I thought it meant they could play at least once and checked for playing and rejected if they did not play with every other player.
What does it mean by "Must reject an unknown player: True should equal False"
Do the playeres have to be sequential from "A" to "Z" or is there characters other than alpha tested for and those should be rejected?
This comment is hidden because it contains spoiler information about the solution
haha, no worries, just repond here or msg me if people complain.
I am fairly decent at Fortran, so it shold work with out problems.
Thanks for the approval.
Thanks, approved. I don't know a thing about Fortran, so hopefully it's ok... :)
I agree, some of the Kata's are really easy in some languages and rediculously hard in other langauages.
It is a shame that kata ranks can be language dependent.
Loading more items...