Re: Nim 1.0.6 is out!
For ARC you can download a nightly build here: [https://github.com/nim-lang/nightlies/releases/tag/2020-01-27-devel-a716543](https://github.com/nim-lang/nightlies/releases/tag/2020-01-27-devel-a716543)
Re: Game unlock gui written with gintro
Nice to hear that. For memory leaks or corruption I generally test with --gc:arc, -d:useMalloc and valgind, and generally get some leak reports. Not only for gintro but also for pure Nim programs, see [https://forum.nim-lang.org/t/5839](https://forum.nim-lang.org/t/5839) for example. But I have still to learn how to exactly interpret what valgrind shows us, and when there are Nim related leaks, then Nim devs will care. For GTK related leaks I will care later, for now it is most important that it runs stable. Please let me know if there should be issues for Windows, use the github issue tracker then.
Re: Nim macro help
In `createSlide`, you have 2 arguments with type varargs[string], because you put a comma after `title`. Also `body.sons` is not defined. import macros proc createSlide(title: string, body: varargs[string]): NimNode = result = newEmptyNode() echo "createSlide", title, body proc slideDslImpl(head, body: NimNode): NimNode = if body.kind == nnkIdent: var args: seq[string] for n in body: if n.kind == nnkStrLit: args.add n.strVal result = createSlide(body.strVal, args) macro slide*(head, body: untyped): untyped = result = slideDslImpl(head, body) echo result.treeRepr # let us inspect the result Run Documentation of the macros module: [https://nim-lang.org/docs/macros.html](https://nim-lang.org/docs/macros.html)
Re: Nim 1.0.6 is out!
> I don’t understand why the next version will be 1.2 whereas the version 1.1.1 > hasn’t been published officially. Do you understand why the current release is 1.0.6 (the previous was 1.0.4), "whereas the version 1.0.5 hasn’t been published officially"?
Re: Nim 1.0.6 is out!
`devel` is never published as an `stable` version by definition.
Re: Game unlock gui written with gintro
I've build the gui and then the whole project with \--gc:arc. On linux everything works fine. No memory leaks or something else. On windows I cannot build because of a dependency (winim). I think I've to wait until \--gc:arc is more stable/more used.
Re: Nim 1.0.6 is out!
I don’t understand why the next version will be 1.2 whereas the version 1.1.1 hasn’t been published officially.
Re: Nim 1.0.6 is out!
> I think it's headed for the 1.1 release. Minor fix: Current devel is already v1.1.1, the next officially released feature release will be 1.2.0.
Re: Nim 1.0.6 is out!
I think it's headed for the 1.1 release. Patch releases never get new features.
Re: Nim 1.0.6 is out!
Does this include the new --gc:arc option? I guess not?
Re: Fizzbuzz game
I have one in Arraymancer that uses a neural network trained on Fizzbuzz on
number between 101 and 1024 and then tested on 1 ..< 100.
It seems to have learned division:
[https://github.com/mratsim/Arraymancer/blob/v0.6.0/examples/ex04_fizzbuzz_interview_cheatsheet.nim](https://github.com/mratsim/Arraymancer/blob/v0.6.0/examples/ex04_fizzbuzz_interview_cheatsheet.nim)
# A port to Arraymancer of Joel Grus hilarious FizzBuzz in Tensorflow:
# http://joelgrus.com/2016/05/23/fizz-buzz-in-tensorflow/
# Interviewer: Welcome, can I get you a coffee or anything? Do you need a
break?
# ...
# Interviewer: OK, so I need you to print the numbers from 1 to 100,
# except that if the number is divisible by 3 print "fizz",
# if it's divisible by 5 print "buzz", and if it's divisible
by 15 print "fizzbuzz".
# Let's start with standard imports
import ../src/arraymancer, math, strformat
# We want to input a number and output the correct "fizzbuzz" representation
# ideally the input is a represented by a vector of real values between 0
and 1
# One way to do that is by using the binary representation of number
func binary_encode(i: int, num_digits: int): Tensor[float32] =
result = newTensor[float32](1, num_digits)
for d in 0 ..< num_digits:
result[0, d] = float32(i shr d and 1)
# For the input, we distinguish 4 cases: nothing, fizz, buzz and fizzbuzz.
func fizz_buzz_encode(i: int): int =
if i mod 15 == 0: return 3 # fizzbuzz
elif i mod 5 == 0: return 2 # buzz
elif i mod 3 == 0: return 1 # fizz
else : return 0
# Next, let's generate training data, we don't want to train on 1..100,
that's our test values
# We can't tell the neural net the truth values it must discover the logic
by itself.
# so we use values between 101 and 1024 (2^10)
const NumDigits = 10
var x_train = newTensor[float32](2^NumDigits - 101, NumDigits)
var y_train = newTensor[int](2^NumDigits - 101)
for i in 101 ..< 2^NumDigits:
x_train[i - 101, _] = binary_encode(i, NumDigits)
y_train[i - 101] = fizz_buzz_encode(i)
# How many neurons do we need to change a light bulb, sorry do a division?
let's pick ...
const NumHidden = 100
# Let's setup our neural network context, variables and model
let
ctx = newContext Tensor[float32]
X = ctx.variable x_train
network ctx, FizzBuzzNet:
layers:
hidden: Linear(NumDigits, NumHidden)
output: Linear(NumHidden, 4)
forward x:
x.hidden.relu.output
let model = ctx.init(FizzBuzzNet)
let optim = model.optimizerSGD(0.05'f32)
func fizz_buzz(i: int, prediction: int): string =
[$i, "fizz", "buzz", "fizzbuzz"][prediction]
# Phew, finally ready to train, let's pick the batch size and number of
epochs
const BatchSize = 128
const Epochs= 2500
# And let's start training the network
for epoch in 0 ..< Epochs:
# Here I should probably shuffle the input data.
for start_batch in countup(0, x_train.shape[0]-1, BatchSize):
# Pick the minibatch
let end_batch = min(x_train.shape[0]-1, start_batch + BatchSize)
let X_batch = X[start_batch ..< end_batch, _]
let target = y_train[start_batch ..< end_batch]
# Go through the model
let clf = model.forward(X_batch)
# Go through our cost function
let loss = clf.sparse_softmax_cross_entropy(target)
# Backpropagate the errors and let the optimizer fix them.
loss.backprop()
optim.update()
# Let's see how we fare:
ctx.no_grad_mode:
echo &"\nEpoch #{epoch} done. Testing accuracy"
let y_pred = model
.forward(X)
.value
.softmax
.argmax(axis = 1)
.squeeze
let score = y_pred.accuracy_score(y_train)
echo &"Accuracy: {score:.3f}%"
echo "\n"
# Our network is trained, let's see if it's well behaved
# Now let's use what we really want to fizzbuzz, numbers from 1 to 100
var x_buzz = newTensor[float32](100, NumDigits)
for i in 1 .. 100:
x_buzz[i - 1, _] = binary_encode(i, NumDigits)
# Wrap them for neural net
let X_buzz = ctx.variable x_buzz
# Pass it through the network
ctx.no_grad_mode:
let y_buzz = model
.forward(X_buzz)
.value
.softmax
.argmax(axis = 1)
.squeeze
# Extract the answer
var answer: seq[string] = @[]
for i in 1..100:
answer.add fizz_buzz(i, y_buzz[i - 1])
echo answer
# @["1", "fizzbuzz", "fizz", "4", "buzz",
Re: Code golfing in Nim
> You can replace indentation with ; and parentheses Now we can just admit that Nim is a syntax skin over Lisp
Re: runnableExamples Question
Yeah that's a local hack (that works).. the back is in the associated closed PR. The real fix is pending because I don't know yet how to add the `--backend` switch just to the `doc` subcommand.
Nim 1.0.6 is out!
Details at [https://nim-lang.org/blog/2020/01/24/version-106-released.html](https://nim-lang.org/blog/2020/01/24/version-106-released.html)
Re: runnableExamples Question
Ah, cool, yes I saw you tried to use a `-d:doccpp` switch in your std_vector repo, but I did not find the issue on Github. Thanks a lot! KR Axel
Re: runnableExamples Question
I needed this feature few weeks back and opened this issue: [https://github.com/nim-lang/Nim/issues/13129](https://github.com/nim-lang/Nim/issues/13129). It's on my list to try to implement the suggested `-backend` switch for `nim doc` that uses the specified backend for compiling and running the runnableExamples and :test: code block examples.
Re: Game unlock gui written with gintro
I have just shipped gintro v0.7.0 with \--gc:arc support, so you may test your app with it. See top of readme in [https://github.com/StefanSalewski/gintro](https://github.com/StefanSalewski/gintro) for details. There are some serious changes, so it may not work for you out of the box. But we should be able to get it working.
Re: Nim macro help
`import macros`
Nim macro help
Hi, I'm very noob with macros, I'm trying to learn, but I'm getting this error: undeclared field: 'kind' for type system.NimNode proc createSlide(title, body: varargs[string]) = echo "createSlide", title, body proc slideDslImpl(head, body: NimNode): NimNode = if body.kind == nnkIdent: var args: seq[string] for n in body.sons: if n.kind == nnkStrLit: args.add n.strVal result = createSlide(body.strVal, args) macro slide*(head, body: untyped): untyped = result = slideDslImpl(head, body) echo result.treeRepr # let us inspect the result Run
runnableExamples Question
Hi, I would like to use `runnableExamples` in a module that needs to be compiled in cpp mode and that needs to be linked against several C++ shared object files. Currently the `nim doc` command that I implemented in the `nimble` file fails because it is not aware of these requirements. Is there a way to handle this? Many thanks in advance. Kind regards, Axel
