After trying the MNIST mlp.py example, I wanted to use my own data to train an MLP. So I load input from *.txt files in python as numpy.ndarray, which are then taken as arguments of model.fit(). But I keep getting the following error:

Traceback (most recent call last):
File "test.py", line 36, in
model.fit(X=train_data, y=train_label, eval_data=(val_data, val_label))
File "/home/code/xxx/mxnet/mxnet_master/python/mxnet/model.py", line 675, in fit
logger=logger)
File "/home/code/xxx/mxnet/mxnet_master/python/mxnet/model.py", line 273, in _train_multi_device
label[islice].copyto(target)
File "/home/code/xxx/mxnet/mxnet_master/python/mxnet/ndarray.py", line 320, in copyto
return NDArray._copyto(self, out=other)
File "/home/code/xxx/mxnet/mxnet_master/python/mxnet/ndarray.py", line 640, in generic_ndarray_function
c_array(NDArrayHandle, [v.handle for v in mutate_vars])))
File "/home/code/xxx/mxnet/mxnet_master/python/mxnet/base.py", line 72, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: [21:58:27] src/ndarray/ndarray.cc:158: Check failed: from.shape() == to->shape() operands shape mismatch

My data is represented as 1500-dim feature vectors. The shapes of my data array are:

train_data.shape = (1800, 1500)
train_label.shape = (1800, 1)
val_data.shape = (200, 1500)
val_label.shape = (200, 1)

The MLP has two output units for the last layer. And I have tried transposing the input data array, using 1500 units for fc1, and using 2-dim labels (for two output units). None of them helps.

So what is the correct way to use my own data? Thanks.