I'm using CentOS 5.5 (Python 3.5.1, gcc 4.8.1). The data_test has 40 samples in it, but model.predict(data_test) gives 60 predictions. Specifically, while mx.io.CSVIter is loading data_test.csv file, it first gets an array filled with 0 like [[ 0. 0. ... 0.]], then it loads no.21 item in csv file through to no.40 item, then from no.1 to no.39, finally it loads no.40 item but doesn't predict on it, giving 60 predictions in all. However no such error on Windows 10 (Python 3.4.4). Any suggestions?

Thanks.