Open ferrine opened 1 year ago
Scan and OpFromGraph are a bit too eager in how they handle SharedVariables, even if they wouldn't necessarily be needed as inputs of the inner graph. For Scan at least, this is a bit related to the whole updates machinery, so that RandomVariables update across iterations without users saying so manually...
Works:
Raises an uninformative or fairly informative error
Scan silently passes and aesara.grad fails miserably later if strict=False
Please provide any additional information below.
Versions and main components
pytensor config:
``` floatX ({'float64', 'float16', 'float32'}) Doc: Default floating-point precision for python casts. Note: float16 support is experimental, use at your own risk. Value: float64 warn_float64 ({'ignore', 'pdb', 'warn', 'raise'}) Doc: Do an action when a tensor variable with float64 dtype is created. Value: ignore pickle_test_value (