qp.math.is_tracer¶
- is_tracer(tensor, like=None)[source]¶
Return whether the tensor is a JAX tracer without a concrete value.
Such tracers commonly occur when tracing functions decorated with
@jax.jit. This function does not identify PennyLane shape/type descriptors such asqp.typing.Float[3]orqp.typing.Wire[3]. Tracers with a concrete value also returnFalse.Note
Currently Autograd tensors and Torch tensors will always return
False. This is because:Autograd does not provide JIT compilation, and
@torch.jit.scriptis not currently compatible with QNodes.
- Parameters:
tensor (tensor_like) – input tensor
like (str) – The name of the interface. Will be determined automatically if not provided.
- Returns:
whether the tensor is a tracer without a concrete value
- Return type:
bool
Example
Consider the following JAX function:
import jax from jax import numpy as jnp def function(x): print("Value:", x) print("Tracer:", qp.math.is_tracer(x)) return jnp.sum(x ** 2)
When we execute it, we see that the tensor is not abstract; it has known value:
>>> x = jnp.array([0.5, 0.1]) >>> function(x) Value: [0.5, 0.1] Tracer: False Array(0.26, dtype=float32)
However, if we use the
@jax.jitdecorator, the tensor will now be abstract:>>> x = jnp.array([0.5, 0.1]) >>> jax.jit(function)(x) Value: Traced<ShapedArray(float32[2])>with<DynamicJaxprTrace(level=0/1)> Tracer: True Array(0.26, dtype=float32)
Note that JAX uses an abstract shaped array, so although we won’t be able to include conditionals within our function that depend on the value of the tensor, we can include conditionals that depend on the shape of the tensor.