Accoding to the statement in the paper, each example is split into several segments with slide window. For instance, the example, whose length is 9000, is supposed to be converted into a matrix [12, 3000], 3000 refers to the size of the window. Each row in the matrix is one segment of the example. In this way, the varied-length example will be converted into matrix with different rows.
If we just put all the rows toghther into one matrtix and send this matrix into a model to train, how could we know which segs belong to which examples? If we dont, how could we make sure that what the model learns is not cross examples?
I checked this code for several times and failed to find the resolution ('test_pid' was not used at all), please help, thanks a lot.
Accoding to the statement in the paper, each example is split into several segments with slide window. For instance, the example, whose length is 9000, is supposed to be converted into a matrix [12, 3000], 3000 refers to the size of the window. Each row in the matrix is one segment of the example. In this way, the varied-length example will be converted into matrix with different rows.
If we just put all the rows toghther into one matrtix and send this matrix into a model to train, how could we know which segs belong to which examples? If we dont, how could we make sure that what the model learns is not cross examples?
I checked this code for several times and failed to find the resolution ('test_pid' was not used at all), please help, thanks a lot.