We should set a random seed eventually, but this will make sure that the scipy implementation here at least matches the theano implementation. I've heard some people do not find theano.scan totally easy to write and understand, so this seems like a worthwhile test to have!
You might take a careful look at how I wrote down the data generating process, and either loosen it up, or tighten it up if it is not a reasonable one!
We should set a random seed eventually, but this will make sure that the scipy implementation here at least matches the theano implementation. I've heard some people do not find theano.scan totally easy to write and understand, so this seems like a worthwhile test to have!
You might take a careful look at how I wrote down the data generating process, and either loosen it up, or tighten it up if it is not a reasonable one!