Thẻ ghi nhớ: AID301c- new | Quizlet

Which of the following is/are NOT native or built-in data types in Python?

A. boolean

B. integer

C. float

D. heap

E. string

F. varchar

DF — D. heap; F. varchar

Mutable data types/collections in Python can be changed in place. Immutable ones can not change in place. Which of the following are mutable?

A. bool

B. int

C. float

D. set

E. list

F. string

G. tuple

H. complex

DE — D. set; E. list

Which of the following is NOT true about Python?

A. Python code can run in IPython and Jupyter notebooks

B. Python allows for the inclusion of comments and pseudocode to better organize code

C. Users can save .py files with an editor then subsequently execute them from the command line

D. Base Python automatically parallelizes processing across cores when multiple cores are available

E. Python allows users to save multiple functions in a .py file then import those functions in a different file

D — D. Base Python automatically parallelizes processing across cores when multiple cores are available

Which of the following pairs of events are mutually exclusive. There can be more than one answer.

A. Odd numbers and the number 3

B. Even numbers and numbers greater than 10

C. Negative numbers and positive numbers less than 25

D. Numbers between 100-200 and numbers between 201-300

E. None of the above

CD — C. Negative numbers and positive numbers less than 25; D. Numbers between 100-200 and numbers between 201-300

If you were to munge the data into a pandas.DataFrame which of the following would describe a reasonable goal for the cleaning process?

A. customer on the rows and items like total_sales, name, most_bought on the columns

B. daily revenue on the rows items like customer_name and total on the columns

C. transactions on the rows and items like customer_name and item_id on the columns

D. None of the above

C — C. transactions on the rows and items like customer_name and item_id on the columns

Which types of programming tasks best describes what you are expected to already have some familiarity with before beginning this course?

A. dashboarding, high performance computing, and code profiling

B. numeric computing, data munging, data visualization and data modeling

C. convex optimization, python programming, statistical programming

D. continuous integration, linear programming, and data exploration

B — B. numeric computing, data munging, data visualization and data modeling

Though the emphasis may change, which two elements are both essential and common to all three process models we talked about?

A. prediction, recommendation

B. data mining, data cleaning

C. resolve the business question, feedback loops

D. testing, model deployment

C — C. resolve the business question, feedback loops

Is the following statement True/False? To succeed in this course you are expected to be proficient in any one of the following: R, Python or Java.

A. TRUE

B. FALSE

B — B. FALSE

Which of the following is the least accurate statement about the advantages of using process models in data science? Process models generally help by...

A. avoiding unnecessary tangents

B. speeding up the process of getting through the workflow

C. minimizing the model selection process

D. guiding effective time allocation

B — B. speeding up the process of getting through the workflow

Is the following statement True/False? Design thinking is applied in other domains which helps make the task of communicating the AI workflow to those outside of data science easier.

A. TRUE

B. FALSE

A — A. TRUE

It is day one on the job and you need to come up with a plan—how do you begin?

A. Gather what data you can quickly and perform some EDA to understand the problem better

B. Plan to interview or study reviews of both satisfied and dissatisfied subscribers as soon as possible

C. Get the perspective from management and follow the leads they might provide

D. Something else entirely

(2 nguon khac dap an!) Nguon 1: B. Plan to interview or study reviews of both satisfied and dissatisfied subscribers as soon as possible / Nguon 2: C. Get the perspective from management and follow the leads they might provide

In order to come up with the back-of-the-envelope ROI calculation for this project, how might you approach it?

A. Number of active users X Yearly payment

B. Number of active users X monthly payment X % increase of users (assumption)

C. Yearly costs X (number of users at month 2 - number of currently active users)

D. Number of active users X Yearly payment - estimate for cost of project time

AD — A. Number of active users X Yearly payment; D. Number of active users X Yearly payment - estimate for cost of project time

Thinking with the lens of the scientific process, what would your next steps be if you wanted to decide where to open the next store for your sled business?

A. Start pulling sales and other data to create a business viability assessment for Vermont

B. Gather more data and repeat the snowfall experiment

C. Gather different data say snowfall by county and repeat the experiment

D. Start a business viability assessment for all three states

ABCD — A. Start pulling sales and other data to create a business viability assessment for Vermont; B. Gather more data and repeat the snowfall experiment; C. Gather different data say snowfall by county and repeat the experiment; D. Start a business viability assessment for all three states

When embarking on a data science project, why do you ultimately want to format your data so that it can be housed in something like a Pandas DataFrame or NumPy Array?

A. DataFrames/Arrays most closely resemble tables in relational databases.

B. DataFrames/Arrays are the only structures in Python capable of holding significant amounts of data.

C. Nearly all modeling algorithms take input data in a tabular format analogous to format of DataFrames/Arrays.

D. All of the above

C — C. Nearly all modeling algorithms take input data in a tabular format analogous to format of DataFrames/Arrays.

Lets imagine there is a start-up that has a speech-to-text service that incorporates gestures and body language into its output. Which of the following products represents the most defensible business opportunity.

A. Offer a service that hooks into streaming video and predicts the emotional state of people in the videos

B. Create an app that allows job interviewers to get additional information about candidates

C. Create a new and improved conferencing app

D. Create a service that improves on existing audio recognition systems as a richer interface to mobile devices

D — D. Create a service that improves on existing audio recognition systems as a richer interface to mobile devices

Lets imagine there is a start-up that has a speech-to-text service that incorporates gestures and body language into its output. They offer annotated meeting reports as a product and customers are generally very satisfied, but sales to new customers tend to be very slow to acquire. Which of the following business opportunities should be the highest priority?

A. Develop and delivery new products to existing customers

B. Develop new products and target new customers

C. Use customer segmentation and/or market analysis to help marketing with new customers

D. Use customer segmentation and/or market analysis to move into a different market

C — C. Use customer segmentation and/or market analysis to help marketing with new customers

Your company is convinced it is time to change the nature of your companies core product and management has come to ask your advise. Which question DOES NOT exemplify scientific thinking in this situation?

A. Do we have any data like a corpus of customer feedback to support this decision?

B. Can we run an experiment like A/B testing to see if it helps support this decision?

C. Which members of leadership support this decision?

D. Have any other companies been successful making a comparable change?

C — C. Which members of leadership support this decision?

Is the following statement true or false? CSV files are one of the most commonly used file formats for data science because file input/output is easy they are plain-text, and they work well with commonly used spreadsheet tools.

A. TRUE

B. FALSE

A — A. TRUE

Which of the following DOES NOT represent a valid relational database to connector relationship?

A. MySQL --> MySQL-python

B. PostgreSQL --> psycopg/psycopg2

C. SQLite --> sqlite3

D. Berkeley DB --> bsddb

D — D. Berkeley DB --> bsddb

Which tasks should be included in a data ingestion pipeline? (Choose one or more)

A. Account for missing data, faulty data, repeated observations and other data integrity issues

B. Ensure that expected data is returned given a specific set of parameters

C. Ensure that an expected format is returned

D. Ensure that models produce expected results

ABC — A. Account for missing data, faulty data, repeated observations and other data integrity issues; B. Ensure that expected data is returned given a specific set of parameters; C. Ensure that an expected format is returned

Sparse matrices can be useful as a target destination for ETL, but what are the main caveats (choose one or more)?

A. You cannot convert directly from a numpy.array to any of the scipy.sparse matrices

B. NumPy linear algebra functions generally cannot be called directly

C. Saving to disk is not possible directly from a scipy.sparse format

D. The train test splits need to be performed by hand with scipy.sparse matrices

E. It is difficult to print to screen scipy.sparse matrices directly

BE — B. NumPy linear algebra functions generally cannot be called directly; E. It is difficult to print to screen scipy.sparse matrices directly

Which types of data generally work well with sparse matrices?

A. word counts, time-series data

B. audio files, images

C. word counts, user-item matrix for recommendations

D. text data, audio files

C — C. word counts, user-item matrix for recommendations

Is the following statement True or False? Sparse matrices from SciPy need to be transformed into a dense matrix before using scikit-learns train-test-split function?

A. TRUE

B. FALSE

B — B. FALSE

Which fundamental part of the data ingestion process is concerned with to the phrase "bad data in equals bad data out"?

A. Gather all relevant data from the sources of provided data

B. Implement several checks for quality assurance

C. Take the initial steps towards automation of the ingestion pipeline

B — B. Implement several checks for quality assurance

Which of the following is most concerned with ensuring deployed models scale well with added users?

A. data scientist

B. data analysts

C. data engineer

D. product manager

C — C. data engineer

Which of the following is statements is the least correct in the context of the EDA process

A. EDA is used to provide summary level insight into a dataset

B. EDA consists of both exploratory and confirmatory data analysis

C. EDA can be used to discover missing data, outliers and class inbalance issues

D. The EDA process can be used to help predict time to completion for a project

E. The EDA process is an ideal time to explore the connection between the data and the business opportunity

B — B. EDA consists of both exploratory and confirmatory data analysis

Which of the following is an example of a data manipulation that is NOT considered reproducible research?

A. Saving classes and functions in a Python file to be called by Jupyter

B. Code blocks in Jupyter notebooks

C. The use of proprietary tools to carry out research

D. Graphics, plots and other visualizations

E. Copy and paste actions in a spreadsheet

E — E. Copy and paste actions in a spreadsheet

True/False. The seaborn pairplot and other seaborn plotting functions exist as distinct tools from the plots available through matplotlib.

A. True

B. False

B — B. False

In the continuing AAVAIL streaming case study example, one of the data features that can be useful in answering questions about customer churn is the total number of streams that a customer has watched. Imagine that you are working with a dataset where 10% of customers are missing this feature. A good place to start would be to go back and see if it's possible to gather this information from the user logs, but assuming that this initiative is unsuccessful, you will have to decide what to do about this missing data. Which course of action is LEAST likely to be helpful in modeling churn?

A. Replace the missing stream count with the mean stream count among users where this information is available.

B. Replace the missing stream count with a -1 to flag that it is unknown for a given user.

C. Use the other features in the dataset in a model to predict the missing stream counts.

B — B. Replace the missing stream count with a -1 to flag that it is unknown for a given user.

What is the main reason for using multiple imputation?

A. Multiple imputation is necessary when more than one feature in the training data has missing values.

B. Multiple imputation is a way to increase the size of your training dataset.

C. Multiple imputation helps to better characterize the error introduced by replacing missing/unknown data with some chosen values.

C — C. Multiple imputation helps to better characterize the error introduced by replacing missing/unknown data with some chosen values.

Which of the following is NOT normally a part of the EDA process

A. Visual summaries of the data

B. Connecting the data to the business opportunity

C. Communication to stakeholders

D. Predictive linear or logistic regression

D — D. Predictive linear or logistic regression

True/False. The EDA process is decoupled from modeling and cannot be used to help estimate the time it will take to complete a modeling procedure.

A. True

B. False

B — B. False

True/False. The software engineering best practice of saving a maximum amount of code in text files for management under version control has become the norm in data science

A. True

B. False

A — A. True

The three types of missingss discussed during this module were:

A. MRAR, MAR, MCAR

B. MNAR, MRAR, MCAR

C. MNAR, MAR, MARC

D. MAR, MRAR, MCAR

E. MCAR, MNAR, MAR

E — E. MCAR, MNAR, MAR

Which statement is the least true about using Jupyter notebooks in the context of EDA

A. They naturally lend themselves to version control systems

B. They can be ported from one environment to another easily

C. They are helpful because a mixture of code and markdown enables storytelling

D. They are integrated with the plotting library matplotlib

E. They are integrated with the data manipulation library pandas

A — A. They naturally lend themselves to version control systems

Which of the following is NOT an example of assumption that you work with when making probability statements about a sample of data?

A. That there is an underlying population that your sample comes from

B. That the population follows an assumed probability distribution

C. That the observations in your sample are independent and identically distributed

D. That random variables represent the possible values that the data can take

E. That the probability statement applies to one observation at a time in a data set

E — E. That the probability statement applies to one observation at a time in a data set

There are many ways to carry out statistical inference. Which one method of the following is NOT used to compute estimates in the context of statistical inference.

A. Null Hypothesis Significance Testing (NHST)

B. Maximum Likelihood Estimation (MLE)

C. Markov Chain Monte Carlo (MCMC)

D. Expectation Maximization (EM)

E. Simulation via Permutations

A — A. Null Hypothesis Significance Testing (NHST)

Company Z sent out a user satisfaction survey to its customers that included some demographic questions. They want to determine if there is a difference in the age among users of Product 1 versus users of Product 2 (at least among the survey respondents). Which of the following is an appropriate null hypothesis for this study?

A. Users of Product 1 are on average older than users of Product 2.

B. Users of Product 1 are on average younger than users of Product 2.

C. Users of Product 1 and Product 2 are on average the same age.

C — C. Users of Product 1 and Product 2 are on average the same age.

Which of the following is the least valid statement when it comes to dashboards?

A. Dashboards are an easy way to share summaries and findings

B. Dashboards have interactive functionality that helps create a rich experience for the user

C. Dashboards are generally used after serveral iterations of the AI workflow

D. Dashboards are quick way to create portable simple plots

E. Dashboards can be used to tell the story of investigative visualizations

D — D. Dashboards are quick way to create portable simple plots

A data scientist at Company Z sorted the survey responses by whether the respondents used Product 1 or Product 2 and then compiled their ages (p1_ages and p2_ages lists). Of the hypothesis tests discussed in these contents what one is the most appropriate for testing the following hypothesis? There is no age difference, on average, between the users of product 1 and the users of product 2

A. A 1-sample t-test

B. A 2-sample t-test assuming equal variance

C. Z-Test with continuity correction

D. A 2-sample unequal variances t-test

E. Binomial

D — D. A 2-sample unequal variances t-test

Suppose that on average 2.5% of visitors to your website sign up for your newsletter. In a recent week, 2701 visitors out of a total of 108879 signed up. Using a binomial distribution, what is the probability that number of visitors who signed up is 2701 or fewer?

A. 0.125

B. 0.346

C. 0.414

D. 0.007

E. 0.015

B — B. 0.346

True/False. If there customer churn were quantified using a Poisson distribution, then a bootstrap could be used to quantify the uncertainty associated with the estimate.

A. True

B. False

A — A. True

Which of the following is NOT and example of a valid strategy to deal with the multiple comparisons problem?

A. Benjamini/Hochberg correction based on False discovery Rates

B. Create a null distribution using permutations to help provide context

C. Perform all comparisons then only keep the single test that performs the best

D. If appropriate use an alternative modeling framework like generalized linear models

E. Bonferroni Correction

C — C. Perform all comparisons then only keep the single test that performs the best

Which scikit-learn API interface would be used to carry out feature engineering with a domain expert?

A. Transformer

B. Estimator

C. Fit

D. Predict

E. Pipeline

A — A. Transformer

Which variant of SMOTE is most appropriate when you have a mixture of categorical and continuous variables?

A. KMeansSMOTE

B. BorderlineSMOTE

C. SVMSMOTE

D. SMOTENC

E. SMOTE

D — D. SMOTENC

Which of the following is not an example of a technique used for dimensionality reduction technique?

A. Latent Dirichlet allocation

B. Non-negative matrix factorization

C. Singular value decomposition

D. Eigenvalue decomposition

E. K-nearest neighbors

F. Principal Components Analysis

E — E. K-nearest neighbors

When printing the most representative words from each topic what best describes the insight we gain?

A. The top words in each topic correspond to the most frequently used words in the corpus

B. The topics are defined by their representative words and the document is a mixture of these topics

C. Documents have topics and the words describe the corpus

D. The words make up the document and the topics describe the words

E. Topics are latent features and the top words describe the average document

B — B. The topics are defined by their representative words and the document is a mixture of these topics

The .fit_transform method corresponds to which scikit-learn interface(s)? Choose one answer.

A. Transformer, Estimator, Predictor

B. Transformer, Estimator

C. Estimator, Predictor

D. Transformer

E. Transformer, Predictor

B — B. Transformer, Estimator

True/False. A principal reason for emphasizing the use of pipelines in the AI workflow is to have a consistent platform that enables comparison of many variants of the workflow.

A. True

B. False

A — A. True

Which of the following statements describes the best strategy to address class imbalance?

A. If there is a lot of data just use under-sampling otherwise use outlier detection algorithms.

B. Determine the best variant of SMOTE by comparisons and use it.

C. Continue to collect data until you have balanced classes.

D. Use an outlier detection algorithm or SVM instead of a re-sampling technique.

E. Compare re-sampling approaches to a baseline and to detection algorithms.

E — E. Compare re-sampling approaches to a baseline and to detection algorithms.

Which of the following statements is not a feature of the package imbalanced-learn?

A. Has a suite of over and under-sampling methods implemented

B. Works with TensorFlow

C. Has a number of tutorials to work from

D. Has outlier detection algorithm packaged as part of library

E. Works with scikit-learn pipelines

D — D. Has outlier detection algorithm packaged as part of library

Which of the following statements does not describe valid use case for dimensionality reduction?

A. Principal components analysis to process images used in classification.

B. Non-negative matrix factorization to resolve topics from a corpus of words.

C. t-distributed stochastic neighbor embedding to visualize the results of a clustering algorithm.

D. Using an ANOVA to select a subset of features

E. Down-sampling of the majority class

E — E. Down-sampling of the majority class

True/False. tSNE is a reasonable alternative to PCA because it describes a wider variety of structures. However, it is still recommended to use another dimensionality reduction method, like PCA if the number of features is very high.

A. True

B. False

A — A. True

Which statement best describes why visualization of topics can have an impact on the business opportunity?

A. Because sharing with domain experts might enable topic-specific feature engineering

B. Because visual inspection can help choose the number of topics

C. Because we are able to see the top words with each topic

D. Because we are able to see the relative importance of each topic across the corpus

E. Because domain experts can visually inspect the validity of the topics

A — A. Because sharing with domain experts might enable topic-specific feature engineering

For credit applications, which of the following types of patterns would be the most undesirable for a supervised learning algorithm to use?

A. Given salary > 50K and dept < 5K they will repay the loan

B. Given salary > 80K they will repay the loan

C. Given salary > 50K and Age > 25 and they will repay the loan

D. Given salary > 50K and that they reside in a metropolitan (population >100K)

E. Given salary > 50K and the degree name

C — C. Given salary > 50K and Age > 25 and they will repay the loan

Which of the following is not an example of an outlier detection algorithms?

A. Adaptive Synthetic (ADASYN)

B. Elliptic Envelope

C. One Class SVM

D. Isolation Forest

E. Local Outlier Factor

A — A. Adaptive Synthetic (ADASYN)

Which of the following is NOT considered a reasonable metric to choose the number of clusters?

A. Calinski-Harbasz index

B. Inertia or the within cluster sum-of-squares

C. Silhouette Score

D. Davies-Bouldin index

E. Adjusted Rand index

B — B. Inertia or the within cluster sum-of-squares

What is the number of profiles or clusters for this particular dataset?

A. 2-4

B. 4-6

C. 6-8

D. 8-10

E. 10-12

C — C. 6-8

You are asked to build a recommendation engine for new products at an online retailer. Which of the following features is the least likely to be a protected attribute? Recall that a protected attribute is one that may contain privileged and unprivileged classes.

A. age

B. gender identity

C. race

D. religion

E. purchase history

E — E. purchase history

True/False. The API for the reweighting algorithm has a number of custom methods that will require you to consult the documentation to ensure appropriate use.

A. True

B. False

B — B. False

Which outlier detection method is known to work well on high-dimensional data?

A. Random Forests

B. Elliptic Envelope

C. One Class SVM

D. Isolation Forest

E. Local Outlier Factor

E — E. Local Outlier Factor

True/False. In an imbalanced dataset if the minority class represents less than 5% of all of the samples then outlier detection algorithms must be used in place of other supervised learning algorithms.

A. True

B. False

B — B. False

Which clustering method can be readily applied to graphs?

A. Gaussian mixture models

B. Spectral clustering

C. Affinity Propagation

D. k-means

E. Dirichlet process Gaussian mixture models

B — B. Spectral clustering

Which of the following clustering methods does not need to set the number of clusters?

A. Gaussian mixture models

B. Spectral clustering

C. MiniBatch k-means

D. k-means

E. Dirichlet process Gaussian mixture models

E — E. Dirichlet process Gaussian mixture models

True/False. In the clustering case study, the suggested re-sampling methods drove a major improvement in model performance.

A. True

B. False

B — B. False

True/False. In the context of customer profiling and the AAVAIL data set it makes sense to first perform a dimension reduction technique like PCA before running the model through a clustering estimator.

A. True

B. False

B — B. False

In which situation would you most strongly consider MAE over RMSE as a regression metric?

A. where we would like to interpret the error metric in terms of the original units

B. like predicting daily temperature where we expect a small range of values

C. like predicting time to failure for a machine where we expect a long tailed distribution of values

D. like predicting the category or topic associated with a document

E. where we would like to interpret the error metric as a squared version of the original units

C — C. like predicting time to failure for a machine where we expect a long tailed distribution of values

If you have data with a large number of features and you are sure that it will take some time to train and tune the model, which approach is LEAST likely to result in a speed improvement during grid-searching?

A. In your pipeline use variance thresholding to limit the number of features

B. Use the Shuffle and split form of cross-validation

C. Use a randomized grid search form of cross validation

D. Randomly subset the data

E. Use PCA to reduce the dimensionality of the data before training

B — B. Use the Shuffle and split form of cross-validation

Which of the following is not an example of a variant/application of gradient decent that we have covered?

A. batch gradient descent

B. mini-batch gradient descent

C. regularized gradient decent

D. stochastic gradient descent

E. gradient descent applied to regression

C — C. regularized gradient decent

What important feature of the Watson NLU API do we have to ensure so that we can repeat an experiment in the future in a reproducible way?

A. GitHub integration

B. Passing a version argument with each request

C. Each future version of the API is guaranteed to keep the same arguments

D. The NLU can be run locally

E. The exchanged JSON is guaranteed to keep the same format

B — B. Passing a version argument with each request

Processing the corpus with the provided lemmatize_document reduces the total number of tokens to what percentage of the original?

A. 10-15%

B. 20-35%

C. 45-50%

D. 70-75%

E. 85-95%

C — C. 45-50%

If we have a situation where false positive is not as potentially costly as a false negative say flagging comments for manual review based on suspected unlawful activity, which of the following is the best approach to consider?

A. Only look at the f-score of the negative class for evaluation

B. Use recall as the evaluation metric

C. Use precision as the evaluation metric

D. Set beta to 0.5 in the fscore

E. Set beta to 2.0 in the fscore

E — E. Set beta to 2.0 in the fscore

True/False. All classifiers in scikit-learn do multi-class classification out-of-the-box. These classifiers can differ in their approach though (e.g one-vs-all or one-vs-one).

A. True

B. False

A — A. True

Which of the following is not an example of a generalized linear model (GLM)?

A. ANOVA

B. Multinomial regression

C. Poisson regression

D. KNN regression

E. Logistic regression

D — D. KNN regression

True/False. Models in the generalized linear mixture model (GLMM) family like multilevel models are generally optimized using sophisticated techniques like MCMC sampling.

A. True

B. False

A — A. True

When you use Watson Services like Watson Natural Language Understanding via the Python SDK, what are the three items that need to be saved? These items are generally saved on a local machine and included in scripts and notebooks as imported variables.

A. service version, service API key, service JSON map

B. service URL, service JSON map, service API key

C. service API key, service version, service URL

D. service version, service IAMAuthenticator, service URL

E. service API key, service URL, service IAMAuthenticator

C — C. service API key, service version, service URL

Which of the following does not describe a feature of the Watson Natural Language Understanding service?

A. Perform document classification tasks using a custom model built from text

B. Identify high-level concepts that aren't necessarily directly referenced in the text

C. Find people, places, events, and other types of entities mentioned in your content

D. Recognize when two entities are related, and identify the type of relation

E. Analyze the sentiment toward specific target phrases and the sentiment of the document as a whole

A — A. Perform document classification tasks using a custom model built from text

Which of the following is not an example of a relevant question when tuning a NLP classification pipeline?

A. Should I use bag-of-words or a vector embedding representation?

B. Which stop words do I include?

C. Which n-grams do I include?

D. Should I use a TF or a tf-idf transformation?

E. Should I use RMSE or MAE as an evaluation metric?

E — E. Should I use RMSE or MAE as an evaluation metric?

Which of the following model is not an example of an ensemble approach to learning?

A. Decision tree

B. Random forest

C. Boosting

D. Model stacking

E. Gradient boosting

A — A. Decision tree

Which of the following neural network architectures are most-commonly used for time-series analysis?

A. Multi-layer perceptron

B. Recurrent neural networks

C. Transfer learning

D. Convolutional neural network

E. Autoencoders

B — B. Recurrent neural networks

True/False. All images must be downloaded and saved locally before you can call classify from the connected IBM Watson Visual Recognition service.

A. True

B. False

B — B. False

A simple CNN that runs on all ten classes and uses all of the data obtains approximately what level of accuracy?

A. 62-77%

B. 78-83%

C. 84-89%

D. 90-94%

E. 95-99%

D — D. 90-94%

True/False. Bagging and boosting ensemble methods both use only decision trees as base classifiers. The difference is in the bias and variance of the individual trees.

A. True

B. False

B — B. False

True/False. A decision tree classifier is useful as a model for the AAVAIL subscriber churn data.

A. True

B. False

A — A. True

Which of the following was not discussed as a tunable parameter of a neural network?

A. Hardware availability

B. Activation functions: sigmoid, tanh, softmax, ReLU, leaky ReLU

C. Regularization techniques: weight decay, early stopping, dropout

D. Training method: Loss function, learning rate, batch size, number of epochs

E. Structure: the number of hidden layers, the number of nodes in each layer

A — A. Hardware availability

True/False. Transfer learning is a recent advancement to come out of the field of reinforcement learning.

A. True

B. False

B — B. False

When training a custom classifier in Watson Visual Recognition the negative images should be:

A. As visually similar as possible to the positive images

B. Background images without the positive images

C. As random as possible to establish a background

D. Randomly generated from the positive images

E. Visually distinct from the positive images

A — A. As visually similar as possible to the positive images

True/False. The Watson Visual Recognition service can only be accessed using an API key via Python or curl.

A. True

B. False

B — B. False

Which of the following use cases is the least appropriate use case for a convolutional neural network?

A. Image classification

B. Image retrieval

C. Image composition

D. Object detection

E. Image segmentation

C — C. Image composition

True/False. A typical convolutional neural network is constructed using a combination of convolutional, pooling and dense layers.

A. True

B. False

A — A. True

True/False. If we continue to add GPUs or other computational resources, the time it takes to train a model will always continue to decrease.

A. True

B. False

B — B. False

True/False. It is reasonable to think of the commands in a Dockerfile as a step-by-step recipe on how to build up a Docker image.

A. True

B. False

A — A. True

What are the two steps that must be carried out if you want to iterate locally on a model then deploy it to the Watson Machine Learning (WML) service?

A. Provision a WML service & Dockerize your model

B. Create a Python virtual environment & Dockerize your model

C. Provision cloud storage & Provision a WML service

D. Provision a WML service & create a Python virtual environment

E. Dockerize your model & Provision cloud storage

D — D. Provision a WML service & create a Python virtual environment

Which of the following lists contains one or more references to a technology that is not a specific python package used to speed up and improve the performance of python code?

A. py-cuda, Cython

B. multiprocessing, mpi4py

C. subprocessing, symmetric-multiprocessing

D. threading, ipyparallel

C — C. subprocessing, symmetric-multiprocessing

A Spark cluster is generally managed using a Docker container and a YAML file.

A. True

B. False

B — B. False

Which of the following is least likely to be a use case for Docker containers?

A. Microservices: many loosely coupled and independently deployable services

B. DevOps/Data Engineers use containers as a common platform for many teams

C. Avoid install overhead with a Spark environment

D. Hybrid, multi-cloud portability for a machine learning model

E. Replacement for a standard virtual machine

E — E. Replacement for a standard virtual machine

Docker images are the basis of containers. It is possible to pull an image from the registry and ask the Docker client to run a container based on that image. Some images are official while many others are user defined.

A. True

B. False

A — A. True

What is the principal reason to create a virtual environment before creating your model locally?

A. Because all models should have their own virtual environment

B. Because it will ensure that the most recent packages are used

C. Because virtual environments can be containerized easily

D. Because the Python client for Watson Machine Learning has specific dependencies

E. Because it ensures that locally created/trained models are compatible with Watson Machine Learning

E — E. Because it ensures that locally created/trained models are compatible with Watson Machine Learning

True/False. You may only pass a pickle file to save your model in the your Watson Machine Learning library.

A. True

B. False

B — B. False

True/False. The Spark ML API uses DataFrames from Spark SQL. They can hold a variety of data types with different columns for storing text including: feature vectors, truth labels, and predictions.

A. True

B. False

A — A. True