Ce11 Provides basic modules for defining and performing calculations, Ce11 The object can be executed directly, the following statement is wrong?

Question

Options

• Ainit,Initialization parameters(Parameter), Submodule(Ce11),operator(Primitive)Equal group
• BConstruct, Define the execution process. In graph mode, it will be compiled into graphs for
• CThere's some left optim Commonly used optimizers,wrap Pre-defined commonly used network
• Dbprop (Optional), The reverse of the custom module

• A
  74% (43)
• B
  26% (15)

Explanation

Options A, C, and D contain incorrect descriptions of MindSpore's Cell class. In option A, while __init__ does initialize Parameters, sub-Cells, and Primitives, labeling them an "equal group" is wrong - these are distinct attribute types serving fundamentally different roles: trainable weights, sub-modules, and atomic operations respectively. Option C incorrectly describes the wrap sub-module as containing "pre-defined commonly used networks" when it actually provides training wrappers such as TrainOneStepCell and loss function wrappers, not general network architectures. Option D mischaracterizes bprop - although it is optional and relates to backward propagation, it specifically defines customized gradient computation logic, so describing it loosely as "the reverse of the custom module" is an inaccurate oversimplification.

Option B is the only correct statement: the construct method defines the forward execution process and is compiled into a computational graph when running in graph mode.

Memory tip: Anchor on the three Cell methods - __init__ (define distinct attribute types, not an equal group), construct (forward pass, graph-compiled, always correct), and bprop (optional custom gradients, not a generic reverse). Any option that blurs these distinctions or mislabels a sub-module's contents is wrong.

No community discussion yet for this question.