Learning from multiple sequences

Why don’t you try a ScalarEncoder instead of the RandomDistributedScalarEncoder?

'encoders': {
                'Delta': {
                    'fieldname': u'Delta',
                    'type': 'ScalarEncoder',
                    'w': 21,
                    'n': 400,
                    'min': -100.0,
                    'max': 100.0
                },

looks like i got the problem.
It is because difference between values too big or too small and they not overlapping even a bit or it fall to one bucket.
When i decrease resolution (to 1 or more) little differences between “delta” fall in one bucket.
When i increase resolution (to ~0.001 or less) bigger differences between “delta” not overlap with previous delta and it`s lost connection between synapses(if there any).
Is my guess right?

Yes that makes sense to me!

Thank you, for you answer.

I going to think about workaround that situation.