Feature Engineering

Question 1

What aspects are in the BigQuery preprocessing?

Answer 1

• representation transformation
• feature construction.

Question 2

What Feature representation includes?

Answer 2

• converting a numeric feature to a categorical feature ( bucketization)
• converting categorical features to a numeric representation (one-hot encoding, learning with counts, sparse feature embeddings and so on)
• Some models work only with numeric or categorical features,
• others can handle mixed type features

Question 3

What Feature construction includes?

Answer 3

• creating new features, either by using typical techniques such as polynomial expansion by using univariate mathematical functions or feature crossing to capture feature interactions
• by using business logic from the domain of the ML use case.

Question 4

What types of two types of feature preprocessing BigQuery ML support?

Answer 4

automatic and manual

Question 5

When BigQuery ML Automatic preprocessing occur?

Answer 5

Automatic preprocessing occurs during training.

Question 6

BigQuery ML provides the transform clause for you to define custom preprocessing using the manual preprocessing functions.

Answer 6

True

Question 7

How is BigQuery data processing realized?

Answer 7

• Built-in SQL math and data processing functions
• parsing common data formats,
• SQL filtering operations to exclude bogus data from your training examples datasets

Question 8

List some of the advanced feature engineering preprocessing functions in BigQuery.

Answer 8

• ML. ML.FEATURE_CROSS(STRUCT(features)) does a feature cross of all the combinations.
• TRANSFORM clause, which allows you to specify all preprocessing during model creation. The preprocessing is automatically applied during the prediction and evaluation phases of machine learning
• ML.BUCKETIZE(f, split_points) where split_points is an array.

Question 9

When does Memorization work in ML?

Answer 9

Memorization works when you have so much data that for any single grid cell within your input space the distribution of data is statistically significant.

Question 10

Feature crosses are about memorization.

Answer 10

True

Question 11

What are the benefits of sparse models?

Answer 11

• Sparse models contain fewer features and therefore are easier to train on limited data.
• Fewer features also means less chance of overfitting.
• Fewer features also mean it is easier to explain to users because only the most meaningful features remain.

Question 12

What is used for windowed functions in BigQuery ML?

Answer 12

For these types of time-windowed features, you will use Beam, batch and streaming data pipelines.

Question 13

The tf.data API enables you to build complex input pipelines from simple, reusable pieces.

Answer 13

True

Question 14

What is a complementary technology to Apache Beam?

Answer 14

• Cloud Dataflow, that is a fully managed platform that allows you to run data processing and feature engineering pipelines at scale.
• Cloud Dataflow is the API for data pipeline building in java or python and Apache Beam is the implementation and execution framework.

Question 15

What do you need to do to implement a data processing pipeline?

Answer 15

To implement a data processing pipeline, you write your code using the Apache Beam APIs and then deploy the code to Cloud Dataflow.

Question 16

Apache Beam SDK comes with a variety of connectors that enable dataflow to read from many data sources, including text files in Google Cloud Storage or file systems, even from real-time streaming data sources, like Google Cloud Pub/Sub or Kafka.

Answer 16

True

Question 17

What are contention issues?

Answer 17

Contention issues are when multiple servers are trying to get a file locked to the same file concurrently

Question 18

What is one key advantage of preprocessing your features using Apache Beam?

Answer 18

The same code you use to preprocess features in training and evaluation can also be used in serving.

Question 19

What is the purpose of a Cloud Dataflow connector?

Answer 19

Connectors allow you to output the results of a pipeline to a specific data sink like Bigtable, Google Cloud Storage, flat file, BigQuery, and more.

Question 20

What are ways you could do feature engineering within TensorFlow?

Answer 20

Within TensorFlow itself using feature columns or by wrapping the feature dictionary and adding arbitrary TensorFlow code.

Question 21

Why do you need to use TensorFlow code when wrapping the feature dictionary?

Answer 21

Because this needs to be code executed as part of the model function that is to be a part of the TensorFlow graph.

Question 22

What is the limit of feature pre-processing in TensorFlow?

Answer 22

The limit here is that we can do pre-processing on a single input only.

Question 23

TensorFlow models - Sequence models are an exception - tend to be stateless.

Answer 23

True

Question 24

How does TensorFlow Transform work?

Answer 24

• With tf.Transform, you are limited to TF methods, but then you get the efficiency of TensorFlow.
• tf.Transform is a hybrid of Apache Beam and TensorFlow.
• tf.Transform is the component used to analyze and transform training dat

Question 25

Dataflow preprocessing only works in the context of a pipeline.

Answer 25

True

Question 26

When do we use TensorFlow vs. DataFlow?

Answer 26

• use Dataflow for preprocessing that needs to maintain state, such as time windows (back-end preprocessing for ML models in Dataflow.).
• use TensorFlow for preprocessing that is based on the provided input only. (on the fly preprocessing for ML models)

Question 27

What are the problems with the typical ML pipeline?

Answer 27

• you need to keep batch and live processing in sync,
• all other tooling, such as evaluation, must also be kept in sync with batch processing.

Question 28

Which technology, Beam or TensorFlow, is better suited to doing on the flight transformation of the input data?

Answer 28

• analysis in Beam (similar to sk-learn fir_transform),
• transform in TensorFlow (similar to sk-transform)

Question 29

What PTransforms are available in tf.Transform?

Answer 29

• AnalyzeAndTransformDataset, which is executed in Beam to create a preprocessed training data set, (similar to sk-learn fir_transform)
• TransformDataset, which is executed in Beam to create the evaluation data set (similar to sk-transform)

Question 30

What does tf.Transform do during the training and serving phase?

Answer 30

Provides a TensorFlow graph for preprocessing

Question 31

Fill in the blank:
The ______________ _______________ is the most important concept of tf.Transform. The ______________ _______________ is a logical description of a transformation of the dataset. The ______________ _______________ accepts and returns a dictionary of tensors, where a tensor means Tensor or 2D SparseTensor.

Answer 31

Preprocessing function