Errro when running Conv model.

[Johnny-dai-git]

Hi,

I am trying to run MNIST datset on Conv1d layer model. Based on my understanding, we will need to faltten the Mnist dataset in to 1D instead of 2D . Actually, the MNIST iterator has the flat label. But it still does not work.

Here is the code I am using:

"""Get the train and the val iterator , Flatten the data from(1,28,28) to (1,784)"""
train = mx.io.MNISTIter(
image = "data/train-images-idx3-ubyte",
label = "data/train-labels-idx1-ubyte",
batch_size = 20,
flat=True,
data_shape = (784, ))

val = mx.io.MNISTIter(
image="data/train-images-idx3-ubyte",
label="data/train-labels-idx1-ubyte",
batch_size=20,
flat=True,
data_shape=(784, ))

"""Here is my small model and I only do the forwarding """

net = nn.Sequential()
net.add(
nn.Conv1D(channels=1,kernel_size=1,in_channels=1),
nn.Dense(units=10)
)
net.initialize()

for i, batch in enumerate(train):

print(batch)
print(i)
data = gluon.utils.split_and_load(batch.data[0].astype("float16"), ctx_list=ctx, batch_axis=0)
label = gluon.utils.split_and_load(batch.label[0].astype("float16"), ctx_list=ctx, batch_axis=0)

with ag.record():
    for x, y in zip(data, label):
        z = net(x)

""" Here is the error message"""
/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/bin/python /Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/souce_code/mxnet/example/gluon/imdb.py
[13:35:18] ../src/io/iter_mnist.cc:110: MNISTIter: load 60000 images, shuffle=1, shape=(20,784)
[13:35:20] ../src/io/iter_mnist.cc:110: MNISTIter: load 60000 images, shuffle=1, shape=(20,784)
DataBatch: data shapes: [(20, 784)] label shapes: [(20,)]
0
Traceback (most recent call last):
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/souce_code/mxnet/example/gluon/imdb.py", line 54, in
z = net(x)
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/gluon/block.py", line 682, in call
out = self.forward(*args)
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/gluon/nn/basic_layers.py", line 55, in forward
x = block(x)
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/gluon/block.py", line 682, in call
out = self.forward(*args)
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/gluon/block.py", line 1258, in forward
return self.hybrid_forward(ndarray, x, *args, **params)
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/gluon/nn/conv_layers.py", line 147, in hybrid_forward
act = getattr(F, self._op_name)(x, weight, bias, name='fwd', **self._kwargs)
File "", line 169, in Convolution
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/_ctypes/ndarray.py", line 82, in _imperative_invoke
check_call(_LIB.MXImperativeInvokeEx(
File "/Users/xiangrenbaibaoxiang/Desktop/johnny_source_code/pcap/conv2d/tmp/venv/lib/python3.9/site-packages/mxnet/base.py", line 246, in check_call
raise get_last_ffi_error()
mxnet.base.MXNetError: Traceback (most recent call last):
File "../src/operator/nn/convolution.cc", line 103
MXNetError: Check failed: dshp.ndim() == 3U (2 vs. 3) : Input data should be 3D in batch-num_filter-x

"""My guess"""

It seems that after flatten, the Minister forget the channels information. after flatten, we only have(batch_size, 1D size), however, they require (batch_size, channel, 1D size). I have several questions:

(1), How can I get the right data from the Mnistiterator ?

(2), Can I reshape the iterator ?

(3). Any other way to figure it out?

Thank you!
Johnny