Introduction to Machine Learning with Sound
Lab 1: Gather and prepare the data
Question 1 :Data gathering is a key component in machine learning.
Question 2 :For machine learning models, data needs to be quantifiable and not comparable.
Question 3 :Audio files can be compared directly.
Question 4 :Truncating audio files to the same length makes them compatible.
Question 5 :Silence detection can be used to locate the start of a tune or noise even though the noise might already be any number of bars into a tune.
Lab 2: Build a machine learning model
Question 1 :In Watson Studio, which type of project do you create for a machine learning project to make it easy to find associated assets and models?
- Watson tools
- Data science
- Business Analytics
- Streams Designer
Question 2 :Which statement is true for a typical machine learning project in Watson Studio?
- Comes with everything included
- Needs an associated Cloud Object Storage only
- Needs an associated Machine Learning service only
- Needs an associated Spark service only
- Needs Cloud Object Storage, a Machine Learning service, and a Spark service
Question 3 :What’s the best way to select columns in a machine learning model in Watson Studio?
- Always select all columns as features
- Select a prediction column and all remaining columns as features
- Select a prediction column and select which remaining, sometimes all, columns are features
- Select only a prediction column
Question 4 :What does a 60%, 20%, 20% split of data mean?
- 60% of the data is used to train models, 20% to test the models, and 20% to test for overfitting.
- Checks that more that 60% of the predictions are correct, no more than 20% are incorrect, and no more than 20% are borderline.
- Takes no more than 60 steps to prepare an estimate and no more than 20 steps to determine the certainty.
- 20% of the data is used to train the models, 60% of the data is used to test the models, and 20% to test for overfitting
Question 5 :How does an overfitting model perform?
- Performs well on all data
- Performs well on data that is used to select the model but performs poorly with other data
- Performs badly on all data
- Performs poorly on data that is used to select the model but performs well with other data
Lab 3: Create predictions in a Node-RED application
Question 1 :Node-RED allows you to import and export flows.
Question 2 :In Node-RED, you can install nodes by using which method?
- Deploy option
- The Dashboard
- Manage Palette option
- The Settings option
Question 3 :Which combination of Node-RED nodes are required to inject audio into a flow?
- file inject and microphone
- inject and microphone
- file inject and inject
- file inject, inject, and microphone
Question 4 :You can use the http request node to do which task?
- Encode a data buffer
- Invoke a REST API
- Create a HTTP endpoint
- Deliver an API response
Question 5 :A machine learning classification prediction response contains which items?
- Raw prediction for each class, probability for each class, all classes
- Raw prediction for each class, probability for each class,
- Raw prediction for each class, probability for each class, prediction, prediction class
- Raw prediction for each class, probability for each class, prediction, prediction class, all classes
Lab 4: Create multiclass classification models
Question 1 :When do you use a multiclass classification?
- When your label column contains two distinct categories
- When your label column contains a discrete number of categories
- When your label column contains a large number of values
- When your label column contains different data types
Question 2 :You run machine learning predictions against which type of model or data?
- Deployed models
- Undeployed models
- Generated models
- Selected estimators
Question 3 :What does each Lite Plan instance of the Watson Machine Learning service allow?
- Multiple deployments of only 1 model
- Single deployments of any number of models
- A maximum total of 5 deployments of any number of models
- Any number of deployments of a maximum of 5 models
Question 4 : What is an application prediction?
- The prediction from one deployed model
- An application consolidated prediction from any number of deployed models
- The best prediction from up to five deployed models
- The best and worst prediction from up to five deployed models
Question 5 :An application must consider only the highest probability scoring prediction from a prediction.
Lab 5: Create UIs and integrate visual recognition
Question 2 :One way to train the Watson Visual Recognition service is to feed it positive images of what you want to predict, say, domestic cats, and negative images, say, dogs, lions, birds, and other animals, that you don’t want to predict.
Question 3 :Which type of node is a one-way communication link that can update a web page every time the Machine Learning service makes a prediction?
- http input
- machine learning
Question 4 :If the Node-RED application for Lab 5 processes nine machine learning models, how many machine learning nodes are required?
Question 5 :The Visual Recognition service is simply an API that you can connect to by using http input and output nodes.
Question 1 : If you split your data by 70%, 20%, 10%, which percentage is used for the training data?
Question 2: You can use digital signal processing to create numbers for audio files so that you can compare the audio files and then use these numbers as the basis of a machine learning model.
Question 3 : In which Node-RED node do you set the Mode field to run a prediction?
- Build Payload Values function node
- Hardcoded test node
- Watson Machine Learning (WML) node
- Prediction Columns node
Question 4 :When you create new projects in Watson Studio, a machine learning service is automatically associated with the new project.
Question 5 : The Naïve Bayes estimator does not work with data that contains negative numbers.
Question 6 : In this course, you use Cloud Object Storage to store data files, such as CSV files.
Question 7 : In Lab 3 of this course, you ran a hardcoded prediction test by using Build Payload Values function node. In the code for this function node, why do the columns start with column 2 rather than column 1?
- Because column 2 contains the value that the machine learning model will be predicting
- Because column 1 includes all the data
- Because column 1 is a feature column
- Because column 1, which is not a feature column, is the value that the machine learning model will be predicting
Question 8 : Why is it necessary to deploy the Python Flask digital signal processing application in Lab 3?
- So that you can feed an audio file to the machine learning node in the Node-RED application
- So that you can run predictions in IBM Cloud
- So that your final application for this course can be provided as an API
- So that you can refetch model lists that include sound files
Question 9 : After you deploy a model in Watson Studio, you see a deployment ID. You use this ID to call the predictor in Node-RED or other application.
Question 10 : The Watson Visual Recognition service can be trained to recognize both audio and images.