There are a lot of data generated during the training of the neural network. What mechanism does TensorFlow use to avoid excessive input data?

Question

Options

• AClient
• Bfeed
• Cplaceholder
• Dfetch

• A
  5% (3)
• B
  6% (4)
• C
  73% (46)
• D
  16% (10)

Explanation

TensorFlow uses placeholders (option C) as a mechanism to handle input data efficiently during training. A tf.placeholder defines a node in the computational graph that acts as a reserved slot for data, allowing training data to be fed in batches at runtime rather than loading the entire dataset into memory at once, which prevents memory overload.

Option A (Client) is not a TensorFlow construct for managing input data; it loosely refers to the session-level client-server architecture, which is unrelated to input feeding. Option B (feed) is a distractor because feed_dict is the argument you pass to session.run() to supply values to placeholders - it is the delivery mechanism, not the placeholder itself. Option D (fetch) refers to the output side of session.run(): the tensors you want to retrieve after a forward pass, not the input management layer.

Memory tip: Think of a placeholder as an empty bowl set on the table before a meal. You set the bowl's size and shape upfront (the data type and shape), and fill it with actual food (data) only when you sit down to eat (run the session) - one serving at a time.

No community discussion yet for this question.