I'd like to get involved in updating this project.

Could a list of known tasks be compiled here please.

I just updated my local git repo with your latest commits. When i issue a require 'mxnet' I get the following error:

> require 'mxnet'
LoadError: cannot load such file -- mxnet/lr_scheduler
	from /big/misc/applications/Ruby/extensions/mxnet/mxnet.rb/lib/mxnet.rb:32:in `require'

May you forgot to commit the lr_scheduler file ?

PS: I keep working on spec files

I haven't been able to see other than by dates of commits which version of MXNet this gem requires. Could indication be made.

Reason for needing to know is for the c api compatability. As far as I'm aware there have been significant internal changes to MXNet, including the deprecation of the Symbolic api

Can you add some usage instructions or possibly list some examples?

I didn't have lists of MXNet features to be implemented in mxnet.rb.
But I think we need such lists to make contributions easier.

Part of the coverage lists

• SGD
• Signum
• FTML
• LARS
• LBSGD
• LAMB
• DCASGD
• NAG
• SGLD
• -ccSGD- **NOT IMPLEMENTED (DEPRECATED) Use SGD
• Adam
• AdaGrad
• RMSProp
• AdaDelta
• Ftrl
• Adamax
• Nadam
• Updater

Affects all versions.
Code to reproduce

require 'mxnet'

a = MXNet::NDArray.ones([5,5])
a[0, 0..4] = MXNet::NDArray.array([1,2,3,4,5])

Output:
git/mxnet.rb/lib/mxnet/ndarray.rb:232:in block in []=' IndexError (slicing stop 5 exceeds limit of 5)

Hi,

I'm interested in MXnet support for Ruby, and I was wondering if this project is still alive or if it's been frozen in favor of other efforts toward the implementation of Ruby as API language for MXnet.

Stefano

For the systems which don't have libmxnet.so, I want to provide the way to install libmxnet.so together. I think specifying --with-libmxnet option to gem install command is good for the way.

I tried to fix the following test Error in symbol but couldn't find the root cause. May be in symbol.c

1) MXNet::Symbol.var 
    Failure/Error: sym.send :set_attributes, **attr
    
    TypeError:
      no implicit conversion of Symbol into String
    # ./lib/mxnet/symbol.rb:371:in `set_attributes'
    # ./lib/mxnet/symbol.rb:371:in `var'
    # ./spec/mxnet/symbol_spec.rb:15:in `block (3 levels) in <module:MXNet>'

Part of coverage lists

Activations

• LeakyReLU
• PReLU
• ELU
• SELU
• GELU
• Swish

Layers

• Sequential

• HybridSequential

• Dense

• Dropout

• BatchNorm

• Embedding

• Flatten

• InstanceNorm

• LayerNorm

• GroupNorm

• Lambda

• HybridLambda

• _Conv

• Conv1D

• Conv2D

• Conv3D

• Conv1DTranspose

• Conv2DTranspose

• Conv3DTranspose

• _Pooling

• MaxPool1D

• MaxPool2D

• MaxPool3D

• AvgPool1D

• AvgPool2D

• AvgPool3D

• GlobalMaxPool1D

• GlobalMaxPool2D

• GlobalMaxPool3D

• GlobalAvgPool1D

• GlobalAvgPool2D

• GlobalAvgPool3D

• ReflectionPad2D
  RNN

• GRU Cell

• LSTM Cell

• GRU Layer

• LSTM Layer

Loss

• L2Loss
• L1Loss
• SigmoidBinaryCrossEntropyLoss
• SoftmaxCrossEntropyLoss
• KLDivLoss
• CTCLoss
• HuberLoss
• HingeLoss
• SquaredHingeLoss
• LogisticLoss
• TripletLoss
• PoissonNLLLoss
• CosineEmbeddingLoss
• SDMLLoss

When multiple libraries that rely heavily on CPU time are being utilised within Ruby such as MXNet, they contend for CPU time.
MXNet is designed for parallel processing and hence it does not make sense to keep the engine bound within the GVL.
For best performance the mxnet engine should be initialised outside the context of the Ruby GVL. We rely on the incubator MXNet team to do their parallel processing with threads correctly.

The following both cause the Ruby runtime to crash:

a = MXNet::Symbol.var('a')
a.send(:infer_shape_impl, false) # => crash
a.infer_type # => crash

The cause is access to a NULL/nil value in the respective C functions.

See #29 for a fix.

Hi,

I'm currently trying to implement NDArray.zeros_like (see https://mxnet.incubator.apache.org/api/python/ndarray/ndarray.html#mxnet.ndarray.NDArray.zeros_like )

I tried writing something like

    def self.ones_like(lhs, rhs, *args, **kwargs)
      Ops.zeros_like(lhs, rhs, *args, **kwargs)
    end

But then I get the following error when trying to call the method:

 2) MXNet::NDArray.zeros_like 
     Failure/Error: def #{op_info.func_name}(#{signature.join(', ')})
     
     ArgumentError:
       wrong number of arguments (given 2, expected 0..1)
     # ./lib/mxnet/ndarray/operation_delegator.rb:66:in `zeros_like'
     # ./spec/mxnet/ndarray_spec.rb:340:in `block (3 levels) in <module:MXNet>'

It looks like operation delegation only works form method with one fixed position argument whereas zeros_like has two.
How to work around that ?

Thank you.

The following code causes "no implicit conversion of Array into Integer (TypeError)".

require "mxnet"
a = MXNet::NDArray.new([[1,2,3],[4,5,6]])
p a.shape

Something as simple as the following causes a crash locally, for me:

$ ./bin/console
[1] pry(main)> MXNet::NDArray.new

Invoking MXNet::NDArray.new constructs an invalid instance. Specifically, ndarray_allocate() is called but a handle is never allocated, so future calls to mxnet_ndarray_get_handle() return NULL which breaks code that doesn't deal with this case (including inspecting the resulting instance).

There isn't any obvious use case for MXNet::NDArray.new and the Python API doesn't have an equivalent, so I simply redefined it to raise an error in the following PR: #26

I'm receiving some strange errors when using this gem in conjunction with eventmachine.

I'm rewriting values into a single NDArray instance (reused). Up to 140 writes per second.

Randomly my event loop crashed from mxnet segfault errors. I'm going to push a list of errors in this thread.

These errors appear regardless of my context, so for simplicity, assume these are in CPU context.

1.)
malloc_consolidate(): invalid chunk size

2.) In this error, there are sometimes unicode characters present for attributes as opposed to blank

ruby/2.6.0/bundler/gems/mxnet.rb-edc85f60baa5/lib/mxnet/ndarray/operation_delegator.rb:91:in `imperative_invoke': MXNetError: Cannot find argument '', Possible Arguments: (MXNet::Error)
----------------
scalar : double, optional, default=0
    The scalar value for assignment.
begin : Shape(tuple), required
    starting indices for the slice operation, supports negative indices.
end : Shape(tuple), required
    ending indices for the slice operation, supports negative indices.
step : Shape(tuple), optional, default=[]
    step for the slice operation, supports negative values.
, in operator _slice_assign_scalar(name="", scalar="267.55", step="[1, 1]", end="[181, 1]", ="[180, 0]")

3.) basic_string::_M_construct null not valid

In order to allow true parallel execution by mxnet engine, mxnet library should run within its own Ractor.

The ractor's thread lock should be unlocked to allow the MXNet threaded engine's thread pool to run in parallel.

If a is an instance of MXNet::NDArray, then a.reshape({1 => 2}) causes a crash. The usage of rb_check_convert_type(...) should be replaced with rb_convert_type(...).

See the following PR for a fix and a test: #31

We import Gluon API mainly from Python's implementation.
Make our Gluon API appropriate to Ruby's programming style, we need to collect the rules of API design.

Collected rules

• Do not import function-style constructor from Python

When slicing out of bounds of 1-D NDArray, an exception occurs correctly.

require 'mxnet'

a = MXNet::NDArray.ones([5])
a[0..7]

git/mxnet.rb/lib/mxnet/ndarray.rb:121:in '_slice' MXNet::Error (MXNetError: Check failed: shape_[0] >= end (5 vs. 8) : Slice end index out of range)

When slicing out of bounds of N-D NDArray, the OOB is ignored and as much as possible of the slice is returned.

require 'mxnet'

a = MXNet::NDArray.ones([5,5])
a[0, 0..7]

[1, 1, 1, 1, 1] <MXNet::NDArray 5 @cpu(0)>

require 'mxnet'

a = MXNet::NDArray.ones([5,5])
a[0..7, 2]

[1, 1, 1, 1, 1] <MXNet::NDArray 5 @cpu(0)>

Consistent error behaviour is expected between these methods.

Hi mrkn,

I'm thinking of contributing to mxnet.rb. Before I start doing things On my own I'd like to get a status from you on the things that you believe must be done in priority. Thank you for your guidance

Laurent

https://travis-ci.org/mrkn/mxnet.rb/jobs/426116334

Reference

• https://mxnet.incubator.apache.org/api/python/ndarray/ndarray.html#mxnet.ndarray.NDArray.__getitem__
• https://docs.scipy.org/doc/numpy-1.13.0/reference/arrays.indexing.html#combining-advanced-and-basic-indexing

I was about to write the spec for NDArray#sqrt when I realized that as it is today I can only call:

z = MXNet::NDArray.sqrt(x)
which means NDArray.sqrt (class method)

I guess for #sqrt to work I need to implement

def sqrt
....
end

in NDArray Ruby class, right ?

The current implementation of MXNet:Symbol#set_attributes uses MXSymbolSetAttr for each key-value pair of specified attributes.
It can be rewritten to call MXSymbolSetAttrs.