How Shape Information is Handled by PyTensor#

Currently, information regarding shape is used in the following ways by PyTensor:

To remove computations in the graph when we only want to know the shape, but not the actual value of a variable. This is done with the Op.infer_shape() method.
To generate faster compiled code (e.g. for a 2D convolution).

Example:

>>> import pytensor
>>> x = pytensor.tensor.matrix('x')
>>> f = pytensor.function([x], (x ** 2).shape)
>>> pytensor.dprint(f)
MakeVector{dtype='int64'} [id A] ''   2
 |Shape_i{0} [id B] ''   1
 | |x [id C]
 |Shape_i{1} [id D] ''   0
   |x [id C]

The output of this compiled function does not contain any multiplication or power computations; PyTensor has removed them to compute the shape of the output directly.

PyTensor propagates information about shapes within a graph using specialized Ops and static Type information (see Types).

Specifying Exact Shape#

You can create variables with static shape information as follows:

pytensor.tensor.tensor("float64", shape=(4, 3, 2))

You can also pass shape information directly to some Ops, like RandomVariables

pytensor.tensor.random.normal(size=(7, 3, 5, 5))

You can use the SpecifyShapeOp to add shape information anywhere in the graph. This allows to perform some optimizations. In the following example, this makes it possible to precompute the PyTensor function to a constant.

>>> import pytensor
>>> x = pytensor.tensor.matrix()
>>> x_specify_shape = pytensor.tensor.specify_shape(x, (2, 2))
>>> f = pytensor.function([x], (x_specify_shape ** 2).shape)
>>> pytensor.printing.debugprint(f) 
DeepCopyOp [id A] ''   0
 |TensorConstant{(2,) of 2} [id B]

Problems with Shape inference#

Sometimes this can lead to errors. Consider this example:

>>> import numpy as np
>>> import pytensor
>>> x = pytensor.tensor.matrix('x')
>>> y = pytensor.tensor.matrix('y')
>>> z = pytensor.tensor.join(0, x, y)
>>> xv = np.random.random((5, 4))
>>> yv = np.random.random((3, 3))

>>> f = pytensor.function([x, y], z.shape)
>>> pytensor.printing.debugprint(f) 
MakeVector{dtype='int64'} [id A] ''   4
 |Elemwise{Add}[(0, 0)] [id B] ''   3
 | |Shape_i{0} [id C] ''   2
 | | |x [id D]
 | |Shape_i{0} [id E] ''   1
 |   |y [id F]
 |Shape_i{1} [id G] ''   0
   |x [id D]

>>> f(xv, yv) # DOES NOT RAISE AN ERROR AS SHOULD BE.
array([8, 4])

>>> f = pytensor.function([x,y], z)# Do not take the shape.
>>> pytensor.printing.debugprint(f) 
Join [id A] ''   0
 |TensorConstant{0} [id B]
 |x [id C]
 |y [id D]

>>> f(xv, yv)  
Traceback (most recent call last):
  ...
ValueError: ...

As you can see, when asking only for the shape of some computation (join in the example above), an inferred shape is computed directly, without executing the computation itself (there is no join in the first output or debugprint).

This makes the computation of the shape faster, but it can also hide errors. In this example, the computation of the shape of the output of join is done only based on the first input PyTensor variable, which leads to an error.

This might happen with other Ops such as Elemwise and Dot, for example. Indeed, to perform some optimizations/rewrites (for speed or stability, for instance), PyTensor assumes that the computation is correct and consistent in the first place, as it does here.

You can detect those problems by running the code without this optimization, using the PyTensor flag optimizer_excluding=local_shape_to_shape_i. You can also obtain the same effect by running in the modes FAST_COMPILE or DebugMode.