theano, indexing a large matrix with relatively small value, but got out of boundary

问题

here is part of my theano code for dl4mt (neural machine translation) with line number. The src_positions is a vector of int64, which I printed the result, with the value no more than 16. but when I use src_positions to index attention_mask_, of which the shape is (100, 100). it got index out of bound error.

Here comes the weird part:

1. firstly, the attention_mask_ and gaussian_mask_ are of the same shape.
2. when I use 0.1 * src_positions to index (replace line 5 with commented line 4). line 8 stays unchanged, the program runs well...
3. The weirder part is that, when I replace line 8 with commented line 7, but leave line 5 unchanged, the program still can run!

I'm not sure whether the problem is ... It's really really weird. hope someone can give me some advice.

    1] p_t_s = p_t * sntlens  # n_samples * 1, pt in equation
    2] src_positions = tensor.cast(tensor.floor(p_t_s), 'int64') # (n_samples, 1)
    3] src_positions = src_positions.reshape([src_positions.shape[0], ])
    4] # batch_mask = attention_mask_[tensor.cast(src_positions * 0.1, 'int64')]      # n_sample * maxlen
    5] batch_mask = attention_mask_[src_positions]      # n_sample * maxlen
    6] attn_mask = batch_mask[:, :msk_.shape[0]] * msk_.T      # n_sample * n_timestep 
    7] # batch_gauss_mask = gaussian_mask_[tensor.cast(src_positions * 0.1, 'int64')]      # n_sample * maxlen
    8] batch_gauss_mask = gaussian_mask_[src_positions]   # n_sample * maxlen 
    9] gauss_mask = batch_gauss_mask[:, :msk_.shape[0]] * msk_.T      # n_sample * n_timestep 

回答1:

It seems that the problem occurs based on the src_positions. There won't be any problem sccording to your description. Maybe the src_positions is changed by your codes other than your posted part

来源：https://stackoverflow.com/questions/41052463/theano-indexing-a-large-matrix-with-relatively-small-value-but-got-out-of-boun