Interesting! Really looks like a bug when the last label axis is set to 1.
As a quick turnaround, what about dropping the 1 axis for all data and label in
the csv.iter and perform a reshape within the model architecture?
Something like:
```
iter <- mx.io.CSVIter(data_csv = "data.csv",
data_shape = c(3,3),
label_csv = "label.csv",
label_shape = c(3,3),
batch.size = 2)
data <- mx.symbol.Variable("data")
data <- mx.symbol.reshape(data, shape = c(3,1,3,0))
label <- mx.symbol.Variable("label")
label <- mx.symbol.reshape(label, shape = c(3,3,1,0))
...
loss <- mx.symbol.LinearRegressionOutput(data = model, label = label)
```
[ Full content available at:
https://github.com/apache/incubator-mxnet/issues/12428 ]
This message was relayed via gitbox.apache.org for [email protected]
