python regression metrics not working

[jmschrei]

I have a network which looks like this:

sequence = mx.symbol.Variable( name='data' )

ip1 = InnerProduct( sequence, 250, name="ip1" )
ip2 = InnerProduct( ip1, 100, name="ip2" )
ip3 = InnerProduct( ip2, 50, name="ip3" )
ip4 = mx.symbol.FullyConnected( data=ip3, num_hidden=1, name="y" )
y = mx.symbol.LinearRegressionOutput( data=ip4, name="y" )

train_data  = mx.io.NDArrayIter( X_train, label={ 'y_label' : y_train }, batch_size=batch_size )
eval_data   = mx.io.NDArrayIter( X_test,  label={ 'y_label' : y_test }, batch_size=batch_size )

model = mx.model.FeedForward( ctx=[ mx.gpu(0), mx.gpu(1), mx.gpu(2), mx.gpu(3) ], 
                                 symbol=y,
                                 num_epoch=15,
                                 learning_rate=0.0001,
                                 wd=0.1,
                                 optimizer='adam',
                           )
model.fit( X=train_data,
           eval_data=eval_data,
           eval_metric=mx.metric.RMSE(),
           batch_end_callback=mx.callback.Speedometer(batch_size)
     )

If I feed in y_train and y_test as an array, like I do with classification problems, I get the following error:

Traceback (most recent call last):
  File "ananas.py", line 84, in <module>
    main()
  File "ananas.py", line 73, in main
    batch_end_callback=mx.callback.Speedometer(batch_size)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/model.py", line 655, in fit
    logger=logger, work_load_list=work_load_list, monitor=monitor)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/model.py", line 241, in _train_multi_device
    eval_metric.update(data_batch.label, executor_manager.cpu_output_arrays)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/metric.py", line 84, in update
    assert label.shape == pred.shape
AssertionError

When I check the shapes, they are(256,) and (256,1) when my batch_size is 256. If I reshape my y's to be matrices with 1 column ( y_train = y_train.reshape( y_train.shape[0], 1 ) ), I get the following error:

INFO:root:Auto-select kvstore type = local_update_cpu
INFO:root:Start training with [gpu(0), gpu(1), gpu(2), gpu(3)]
[10:58:18] ./dmlc-core/include/dmlc/logging.h:208: [10:58:18] src/ndarray/ndarray.cc:159: Check failed: from.shape() == to->shape() operands shape mismatch
Traceback (most recent call last):
  File "ananas.py", line 86, in <module>
    main()
  File "ananas.py", line 75, in main
    batch_end_callback=mx.callback.Speedometer(batch_size)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/model.py", line 655, in fit
    logger=logger, work_load_list=work_load_list, monitor=monitor)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/model.py", line 218, in _train_multi_device
    executor_manager.load_data_batch(data_batch)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/executor.py", line 417, in load_data_batch
    _load_label(data_batch, self.label_arrays)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/executor.py", line 301, in _load_label
    _load_general(batch.label, targets)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/executor.py", line 293, in _load_general
    d_src[slice_idx].copyto(d_dst)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/ndarray.py", line 361, in copyto
    return NDArray._copyto(self, out=other)
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/ndarray.py", line 654, in unary_ndarray_function
    c_array(NDArrayHandle, (out.handle,))))
  File "/net/noble/vol1/home/jmschr/anaconda/lib/python2.7/site-packages/mxnet-0.5.0-py2.7.egg/mxnet/base.py", line 76, in check_call
    raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: [10:58:18] src/ndarray/ndarray.cc:159: Check failed: from.shape() == to->shape() operands shape mismatch

How do I run regression?

[piiswrong]

you can fix the RMSE metric (not your input y_train) by reshaping the label or prediction.

[jmschrei]

I did end up reshaping the values in the RMSE metric.