cnevd/deeplearning-mxnet

MXNet for CTR

DeepLearning-MX

• DCNN: A Convolutional Neural Network for Modelling Sentences
• DCPM: Deep Click Prediction Model
• ImgFeatLearn: Image Feature Learning for Cold Start Problem in Display Advertising
• Recomm: Deep Recommended System: SVD
• SNNs: Self-Normalizing Neural Networks

deeplearning-mxnet's Issues

how to made data

how to made data twitter.pkl??

wide convolution question

Hi, it's so kind of you to provide the code of DCNN.
I run it and encounter an error.
It'sFile "D:\Anaconda3\lib\site-packages\mxnet-0.7.0-py3.5.egg\mxnet\base.py", line 77, in check_call raise MXNetError(py_str(_LIB.MXGetLastError())) mxnet.base.MXNetError: Invalid Parameter format for pad expect Shape(tuple) but value='(3.5, 0)'

In this line https://github.com/CNevd/DeepLearning-Mxnet/blob/master/DCNN/dcnn_train.py#L88
the pad=(7/2=3.5, 0), but it should be an int.
And I think the output shape of 1-dim wide convolution should be (s+w-1, xxx) in this paper.
If the input shape of the conv_layer is (batch_size, 1, sentence_size, embed_num), the pad of the convlayer should be (kernel_size-1, 0), instead of (kernel_size/2, 0).
For example,

data = mx.sym.Variable('data')
conv = mx.sym.Convolution(data=data, kernel=(7,1), num_filter=6, pad=(7-1,0))
arg_shapes, output_shape, _ = conv.infer_shape(data=(50,1,100,300))
output_shape
Out[6]: [(50, 6, 106, 300)]

106 = s+w-1 = 100+7-1

Please do not hesitate to correct me if I am wrong.
Many thanks!

k max pooling

have you encountered this?

Traceback (most recent call last):
  File "_ctypes/callbacks.c", line 234, in 'calling callback function'
  File "D:\Anaconda3\lib\site-packages\mxnet-0.7.0-py3.5.egg\mxnet\operator.py", line 600, in creator
    op_prop = prop_cls(**kwargs)
TypeError: __init__() keywords must be strings
Traceback (most recent call last):
  File "D:/DeepLearning-MXNet-master/DCNN/dcnn_train.py", line 196, in <module>
    train_dcnn(args,ctx)
  File "D:/DeepLearning-MXNet-master/DCNN/dcnn_train.py", line 126, in train_dcnn
    dcnn = get_dcnn(sentence_size, args.embed_size, args.batch_size, vocab_size)
  File "D:/DeepLearning-MXNet-master/DCNN/dcnn_train.py", line 97, in get_dcnn
    pool_outi = mx.symbol.Custom(data=fold_outi, name='k_max_pool%s' % i, op_type='k_max_pool', k=ki)
  File "D:\Anaconda3\lib\site-packages\mxnet-0.7.0-py3.5.egg\mxnet\symbol.py", line 1062, in creator
    ctypes.byref(sym_handle)))
OSError: exception: access violation writing 0x000000D9C3076930

seems the k max pooling init failed.
how to solve it?
thanks a lot