Natural Language Processing in TensorFlow Quiz Answers

Week 01: Quiz Answers

Q1. What is the name of the object used to tokenize sentences?

Tokenizer
WordTokenizer
CharacterTokenizer
TextTokenizer

Q2. What is the name of the method used to tokenize a list of sentences?

fit_on_texts(sentences)
tokenize(sentences)
tokenize_on_text(sentences)
fit_to_text(sentences)

Q3. Once you have the corpus tokenized, what’s the method used to encode a list
of sentences to use those tokens?

texts_to_tokens(sentences)
text_to_sequences(sentences)
text_to_tokens(sentences)
texts_to_sequences(sentences)

Q4. When initializing the tokenizer, how to you specify a token to use for unknown words?

out_of_vocab=
unknown_token=
oov_token=
unknown_word=

Question 5: If you don’t use a token for out of vocabulary words, what happens at
encoding?

The word isn’t encoded, and is replaced by a zero in the sequence
The word is replaced by the most common token
The word isn’t encoded, and is skipped in the sequence
The word isn’t encoded, and the sequencing ends

Q6. If you have a number of sequences of different lengths, how do you ensure
that they are understood when fed into a neural network?

Make sure that they are all the same length using the pad_sequences method of the tokenizer
Use the pad_sequences object from the tensorflow.keras.preprocessing.sequence namespace
Specify the input layer of the Neural Network to expect different sizes with dynamic_length
Process them on the input layer of the Neural Netword using the pad_sequences property

Q7. If you have a number of sequences of different length, and call pad_sequences
on them, what’s the default result?

Nothing, they’ll remain unchanged
They’ll get cropped to the length of the shortest sequence
They’ll get padded to the length of the longest sequence by adding zeros to the beginning of shorter ones
They’ll get padded to the length of the longest sequence by adding zeros to the end of shorter
ones

Q8. When padding sequences, if you want the padding to be at the end of the
sequence, how do you do it?

Call the padding method of the pad_sequences object, passing it ‘post’
Pass padding=’post’ to pad_sequences when initializing it
Call the padding method of the pad_sequences object, passing it ‘after’
Pass padding=’after’ to pad_sequences when initializing it

Week 02: Quiz Answers

Q1. What is the name of the TensorFlow library containing common data that you
can use to train and test neural networks?

TensorFlow Datasets
TensorFlow Data
TensorFlow Data Libraries
There is no library of common data sets, you have to use your own

Q2. How many reviews are there in the IMDB dataset and how are they split?

60,000 records, 50/50 train/test split
50,000 records, 80/20 train/test split
60,000 records, 80/20 train/test split
50,000 records, 50/50 train/test split

Q3. How are the labels for the IMDB dataset encoded?

Reviews encoded as a number 1-10
Reviews encoded as a number 0-1
Reviews encoded as a boolean true/false
Reviews encoded as a number 1-5

Q4. What is the purpose of the embedding dimension?

It is the number of dimensions required to encode every word in the corpus
It is the number of words to encode in the embedding
It is the number of dimensions for the vector representing the word encoding
It is the number of letters in the word, denoting the size of the encoding

Q5. When tokenizing a corpus, what does the num_words=n parameter do?

It specifies the maximum number of words to be tokenized, and picks the most common ‘n’ words
It specifies the maximum number of words to be tokenized, and stops tokenizing when it reaches n
It errors out if there are more than n distinct words in the corpus
It specifies the maximum number of words to be tokenized, and picks the first ‘n’ words that were tokenized

Q6. To use word embeddings in TensorFlow, in a sequential layer, what is the
name of the class?

tf.keras.layers.Embedding
tf.keras.layers.WordEmbedding
tf.keras.layers.Word2Vector
tf.keras.layers.Embed

Q7. IMDB Reviews are either positive or negative. What type of loss function

should be used in this scenario?
Categorical crossentropy
Binary crossentropy
Adam
Binary Gradient descent

Q8. When using IMDB Sub Words dataset, our results in classification were poor.
Why?

Sequence becomes much more important when dealing with subwords, but we’re ignoring word positions
The sub words make no sense, so can’t be classified
Our neural network didn’t have enough layers
We didn’t train long enough

Week 03: Quiz Answers

Q1: Why does sequence make a large difference when determining the semantics of
language?

Because the order of words doesn’t matter
Because the order in which words appear dictate their meaning
It doesn’t
Because the order in which words appear dictate their impact on the meaning of the sentence

Q2. How do Recurrent Neural Networks help you understand the impact of
sequence on meaning?

They look at the whole sentence at a time
They shuffle the words evenly
They carry meaning from one cell to the next
They don’t

Q3. How does an LSTM help understand the meaning when words that qualify each
Others aren’t necessarily beside each other in a sentence?

They shuffle the words randomly
They load all words into a cell state
Values from earlier words can be carried to later ones via a cell state
They don’t

Q4. What keras layer type allows LSTMs to look forward and backward in a
sentence?

Bilateral
Unilateral
Bothdirection
Bidirectional

Q5. What’s the output shape of a bidirectional LSTM layer with 64 units?

(128,1)
(128,None)
(None, 64)
(None, 128)

Q6. When stacking LSTMs, how do you instruct an LSTM to feed the next one in the
sequence?

Ensure that return_sequences is set to True only on units that feed to another LSTM
Ensure that return_sequences is set to True on all units
Do nothing, TensorFlow handles this automatically
Ensure that they have the same number of units

Q7. If a sentence has 120 tokens in it, and a Conv1D with 128 filters with a Kernal
size of 5 is passed over it, what’s the output shape?

(None, 120, 128)
(None, 116, 128)
(None, 116, 124)
(None, 120, 124)

Q8. What’s the best way to avoid overfitting in NLP datasets?

Use LSTMs
Use GRUs
Use Conv1D
None of the above

Week 04: Quiz Answers

Q1. What is the name of the method used to tokenize a list of sentences?

tokenize(sentences)
tokenize_on_text(sentences)
fit_on_texts(sentences)
fit_to_text(sentences)

Q2. If a sentence has 120 tokens in it, and a Conv1D with 128 filters with a Kernal
size of 5 is passed over it, what’s the output shape?

(None, 120, 128)
(None, 120, 124)
(None, 116, 128)
(None, 116, 124)

Q3. What is the purpose of the embedding dimension?

It is the number of words to encode in the embedding
It is the number of dimensions required to encode every word in the corpus
It is the number of letters in the word, denoting the size of the encoding
It is the number of dimensions for the vector representing the word encoding

Q4. IMDB Reviews are either positive or negative. What type of loss function should be used in this scenario?

Adam
Binary Gradient descent
Categorical crossentropy
Binary crossentropy

Q5. If you have a number of sequences of different lengths, how do you ensure
that they are understood when fed into a neural network?

Use the pad_sequences object from the tensorflow.keras.preprocessing.sequence namespace
Make sure that they are all the same length using the pad_sequences method of the tokenizer
Process them on the input layer of the Neural Network using the pad_sequences property
Specify the input layer of the Neural Network to expect different sizes with dynamic_length

Q6. When predicting words to generate poetry, the more words predicted the more
likely it will end up gibberish. Why?

It doesn’t, the likelihood of gibberish doesn’t change
Because the probability that each word matches an existing phrase goes down the more words you create
Because the probability of prediction compounds, and thus increases overall
Because you are more likely to hit words not in the training set

Q7. What is a major drawback of word-based training for text generation instead
of character-based generation?

Word based generation is more accurate because there is a larger body of words to draw from
Character based generation is more accurate because there are less characters to predict
There is no major drawback, it’s always better to do word-based training
Because there are far more words in a typical corpus than characters, it is much more memory intensive

Q8. How does an LSTM help understand the meaning when words that qualify each
Others aren’t necessarily beside each other in a sentence?