Cannot find engine branch for minpy package

[mengyuest]

I want to use GPU on numpy opers so I tried minpy. The system is ubuntu16.04 and my display card is GTX1080 (cudnn 7.0). When I installed mxnet, I could not find "engine" branch mentioned in instruction step git clone --recursive -b engine https://github.com/dmlc/mxnet.git
So I just installed the master branch one and at last I could not use minpy correctly. My program wrote as

import minpy.numpy as np
import minpy.numpy.random as random
import time
x= random.rand(1024, 1024)
y= random.rand(1024, 1024)
st = time.time()
for i in range(10):
    z= np.dot(x, y)
z.asnumpy()
print('time: {:.3f}.'.format(time.time()-st))                                                        

and when running, it said
W0123 14:35:58 22806 minpy.dispatch.registry:register:47] Type MXNet for name reshape has already existed
Can anyone tell me what to do? Thanks.

[sxjscience]

I think you may need to raise an issue in dmlc/minpy. Also, some features in minpy like autograd has been merged into MXNet and you can try that. You may also try to write your functions using NDArray APIs instead of relying on minpy.

[ankkhedia]

@mengyuest Did you raise an issue with dmlc/minpy? It seems like a bug with their documentation.
Also it seems from their readme that the repository has been deprecated and merged with Mxnet Gluon. So, did you try to use that?

@sandeep-krishnamurthy Could you please tag it as

• Pending Requester Info
• Gluon

[mengyuest]

@ankkhedia Yeah I have already raised an issue to dmlc/minpy and they told me to use gluon. I tried gluon and it worked. Thanks. Followed is what they told me.

Hi @mengyuest , most minpy's functionality have been merged to Gluon. It supports dynamic autograd and quite extensive numpy operator support. Please have a look and see if it helps!

[sxjscience]

@mengyuest Sounds great! I'll close the issue.

[belomeriem]

Hi,
How did you fix the problem? I am running third-party code and they have import minpy.numpy as np from minpy import core from minpy.nn.solver import Solver which produces the same error you had. I have tried to use Gluon but this does not seem to help. Any suggestions? Thanks

[mengyuest]

I used this and it worked well (though the function is very simple).

import mxnet as mx
from mxnet import nd
import numpy as np
import time

t0=time.time()
mx.random.seed(1)
x=nd.array([[1,2,3,4,5,6],[3,4,56,7,8,9]],ctx=mx.gpu(0))
y1=nd.dot(x,x.T)
print(y1)
t1=time.time()
x=np.array([[1,2,3,4,5,6],[3,4,56,7,8,9]])
y2=np.dot(x,x.T)
print(y2)
t2=time.time()
print(t1-t0,t2-t1)