ENH: modify mrg to not allocate so many streams when computing large random tensors

This (a) wastes memory and (b) wastes a lot of time, because the code for initializing the random state is in Python and sometimes takes for freaking ever.

ENH: modify mrg to not allocate so many streams when computing large random tensors
6341d126 · James Bergstra · 572bf565 · 6341d126
--- a/theano/sandbox/rng_mrg.py
+++ b/theano/sandbox/rng_mrg.py
@@ -618,7 +618,7 @@ def guess_n_streams(size, warn=True):
            r *= s
        if r > 6:
            r = r/6 # chosen as fastest for rbm_benchmark
-        return r
+        return min(r, 30 * 256)
    else:
        if warn:
            warnings.warn((