Free TensorFlow Developer practice test

A. ✓ Correct: Setting the pretrained base model to non-trainable prevents its weights from being updated during feature extraction.

B. × Incorrect: Run_eagerly changes execution style, not whether pretrained layers are frozen.

C. × Incorrect: Classifier activation affects output behavior, not trainability of the base network.

D. × Incorrect: Validation_steps controls evaluation iteration count and does not freeze weights.

A. ✓ Correct: Caching preprocessed data with Dataset.cache improves pipeline performance by avoiding redundant preprocessing across multiple epochs.

B. × Incorrect: Dataset.batch groups elements into fixed-size batches but does not cache them for reuse, thus it doesn't enhance efficiency across epochs.

C. × Incorrect: Dataset.prefetch overlaps preprocessing tasks with model execution to optimize input pipeline performance during training, but it does not cache data.

D. × Incorrect: Dataset.map applies a function to each element of the dataset but does not involve caching preprocessed data for reuse.

A. ✓ Correct: Unfreeze the last few layers and keep others frozen because this approach enables targeted updates to the classifier while maintaining the integrity of earlier feature detectors.

B. × Incorrect: Train all layers simultaneously with a low learning rate because training all layers at once can lead to overfitting due to the large number of parameters being adjusted.

C. × Incorrect: Replace the entire classifier with a new dense layer because replacing the classifier entirely would discard valuable pre-trained features and require extensive retraining.

D. × Incorrect: Freeze all layers except for the first convolutional block because unfreezing only early layers is less effective for fine-tuning tasks that benefit from later feature extraction.

A. ✓ Correct: `from_generator()` is recommended for creating a TensorFlow dataset from a Python generator.

B. × Incorrect: `from_tensor_slices()` creates datasets from NumPy arrays or tensors, not generators.

C. × Incorrect: `make_one_shot_iterator()` is used to create an iterator that reads the entire dataset in one go and does not support dynamic data generation.

D. × Incorrect: `initializable_iterator()` requires manual initialization with a session run call and does not directly work with Python generators.

A. ✓ Correct: The guide explains that a tf.Variable is a tensor whose value can be changed by running ops on it.

B. × Incorrect: This describes initialization only and misses the key point that Variables keep changing during training.

C. × Incorrect: Constants are immutable, while Variables are specifically meant to be updated.

D. × Incorrect: Shape declaration is not the primary purpose of `tf.Variable`.

A. ✓ Correct: The `validation_data` parameter specifies the data to be used for evaluation during training. This allows you to assess model performance on unseen data, which helps prevent overfitting.

B. × Incorrect: The `metrics` parameter defines the metrics that should be calculated during evaluation but does not specify validation data.

C. × Incorrect: The `loss` parameter specifies the loss function used for training and does not relate to validation data.

D. × Incorrect: The `batch_size` parameter controls the number of samples processed before updating model weights, which is unrelated to specifying validation data.

A. ✓ Correct: Bidirectional runs a recurrent layer over the sequence in both directions.

B. × Incorrect: Label conversion is not the role of a recurrent layer wrapper.

C. × Incorrect: Dataset caching is handled by tf.data pipeline methods.

D. × Incorrect: Freezing filters is a transfer-learning image-model concern.

A. ✓ Correct: The largest value in the prediction output identifies the class the model considers most likely.

B. × Incorrect: Averaging scores would lose the per-class ranking needed for classification.

C. × Incorrect: Class_names only maps indices to labels; it does not decide the prediction.

D. × Incorrect: Training labels are not used to classify a new input image.

A. ✓ Correct: Dataset.map applies a function to each element of the dataset, allowing for transformations such as reshaping before feeding into the model.

B. × Incorrect: Dataset.shuffle randomly shuffles elements in the dataset but does not apply a transformation function.

C. × Incorrect: Dataset.batch groups elements from the dataset into batches but does not reshape individual elements.

D. × Incorrect: Dataset.prefetch fetches the next element of an input pipeline ahead of time to improve performance, without applying transformations.

A. ✓ Correct: It correctly identifies the method used to load image datasets from a directory containing subdirectories of images labeled by class name, as per the TensorFlow tutorial.

B. × Incorrect: Tf.data.Dataset.from_tensor_slices() is used for creating datasets from tensor slices and not specifically for loading images from directories.

C. × Incorrect: Tf.io.read_file() reads the contents of a file into a string Tensor but does not load image datasets from directories.

D. × Incorrect: Tf.image.decode_jpeg() decodes JPEG-encoded images to 3-D uint8 tensors and is used for decoding images, not loading them from directories.