CSF 7.7 Assessment - Machine Learning Model Training

Front

Which wombat do you most relate to at this moment?

Back

undefined

Front

Which of the following best describes a corpus in data science? A type of metadata that describes the properties of a dataset, A machine learning model that has been trained on a large dataset, A set of pre-labeled images used to test the accuracy of a machine learning model, A collection of texts or documents used to analyze and extract information

Back

A collection of texts or documents used to analyze and extract information

Answer explanation

A corpus is a large and structured set of texts or documents that is used for natural language processing (NLP) and text analytics in data science. It can be composed of various types of data, including news articles, social media posts, academic papers, or any other written content that can be processed by a computer. The purpose of a corpus is to provide a representative sample of language use in a particular domain, language, or culture, so that researchers and data scientists can analyze it and extract insights about language patterns, semantic relationships, sentiment, or other aspects of language use. In NLP, a corpus is often used for tasks such as training and testing machine learning models, building language models, or developing text classification algorithms.

Front

Which of the following should be true about a corpus used as input for a machine learning model?
It is large, It is sorted, It is labeled, All of these options

Back

It is large

Answer explanation

The statement "It is large" should be true about a corpus used as input for a machine learning model. This is because a larger corpus provides more diverse and representative examples of the language being analyzed, which can improve the accuracy and robustness of the machine learning model.

The statement "It is sorted" is not necessarily true for a corpus used as input for a machine learning model. However, it can be helpful to organize the corpus by categories or topics to facilitate analysis and model training.

The statement "It is labeled" is also not necessarily true for a corpus used as input for a machine learning model, but it can be helpful in supervised learning scenarios where the model is trained on labeled data with known outcomes. However, unsupervised learning techniques can also be used to analyze unstructured text data without the need for labeled data.

Front

Which of the following is NOT an example of metadata? The date a photo was taken, The resolution of an image, The color of a text in a document, The author of an article

Back

The color of a text in a document

Answer explanation

Metadata refers to descriptive information about data that provides context, structure, or meaning to it. In data science, metadata can be used to describe various properties of a dataset, such as its format, structure, source, or content. Metadata can help data scientists to understand and analyze data more effectively, as well as to retrieve, manage, and share data with others.

Examples of metadata include the date a photo was taken, the resolution of an image, and the author of an article. These pieces of information provide context and descriptive information about the photo, image, or article. However, the color of text in a document is not considered metadata, as it is a visual attribute of the text and does not provide information about the document itself.

Front

Which of the following is the best example of metadata? The kind of device that captured the image, Pixels, The differences in images of dogs and wolves, The accuracy of the machine learning model for the image

Back

The kind of device that captured the image

Answer explanation

The best example of metadata from the options given is "The kind of device that captured the image".

Metadata refers to data that provides information about other data. In this case, the type of device that captured an image is metadata because it provides information about the image itself, such as the resolution, image format, and other characteristics that are specific to the device used to capture it.

"Pixels" are not metadata, as they refer to the actual content of the image itself and not information about the image.

"The differences in images of dogs and wolves" is not metadata, as it refers to the content of the images and not information about the images.

"The accuracy of the machine learning model for the image" is also not metadata, as it refers to the performance of the machine learning model on the image and not information about the image itself.

Front

Which of the following is an example of underfit in classification models? Misclassifying images of cats and dogs as horses, Memorizing the training set and failing to generalize to new, unseen data, Correctly classifying images in the training set but performing poorly on the test set, Overfitting the training data and failing to generalize to new, unseen data

Back

Correctly classifying images in the training set but performing poorly on the test set

Answer explanation

Underfitting occurs when a model is too simple to capture the underlying patterns in the data, resulting in poor performance on both the training set and the test set. In classification models, underfitting can occur when the model is not complex enough to distinguish between different classes or when the training data is not representative enough to capture the variability of the data.

Option A) Misclassifying images of cats and dogs as horses is an example of a model that is overfitting the training data and is not generalizing well to new data.

Option B) Correctly classifying images in the training set but performing poorly on the test set is an example of a model that is underfitting the training data and is not capturing the underlying patterns in the data.

Option C) Memorizing the training set and failing to generalize to new, unseen data is an example of a model that is overfitting the training data and is not generalizing well to new data.

Option D) Overfitting the training data and failing to generalize to new, unseen data is also an example of overfitting, similar to option A.

Front

Which of the following is the best example of overfit? Blueberries, raspberries, and cherries being labeled separately instead of all being labeled as “fruit”, Carrots, oranges, and basketballs all being labeled as a pumpkin, Frogs and dogs being mislabeled as the same thing because they both end in “ogs”, A dress being mislabeled as a skirt

Back

Blueberries, raspberries, and cherries being labeled separately instead of all being labeled as “fruit”

Answer explanation

The best example of overfitting from the options given is "Blueberries, raspberries, and cherries being labeled separately instead of all being labeled as 'fruit'".

Overfitting occurs when a machine learning model is trained on a dataset that is too specific or complex, and as a result, it becomes overly tuned to the training data and performs poorly on new or unseen data. In this case, if the machine learning model were trained to identify individual types of berries instead of recognizing them all as "fruit," it would be overfitting to the training data and may not perform well when presented with new types of fruit.

The other examples provided do not illustrate overfitting. "Carrots, oranges, and basketballs all being labeled as a pumpkin" is an example of mislabeling or incorrect labeling, not overfitting. "Frogs and dogs being mislabeled as the same thing because they both end in 'ogs'" is an example of incorrect labeling based on superficial similarities, not overfitting. "A dress being mislabeled as a skirt" is an example of a simple labeling error.

Front

How can data scientists tell if a machine learning model is ready to be used?

Back

Split the available data into a training set and a test set and evaluate the performance on the test set.

Answer explanation

The performance of the machine learning model can be assessed by splitting the data into a training set and a test set. The training set is used to train the model, while the test set is used to evaluate the model's performance on data it has not seen before. By evaluating the model on a separate test set, data scientists can assess how well the model is generalizing to new data, which is a key indicator of whether the model is ready to be used.

The other options are incorrect:

"The model is ready when it has been given exactly 1,000 images as its input corpus" is not a valid criterion for determining whether a model is ready for use, as the appropriate size of the input corpus depends on the specific problem being addressed and the complexity of the model.

"If a model does not perform well, the data scientist has to start all over" is not necessarily true. In many cases, adjusting the model parameters or modifying the training process can improve model performance without requiring a complete restart.

"When the model is underfit, it is ready to be used" is also not true. An underfit model is one that has not learned enough from the training data, and will likely perform poorly on new data. The model needs to be further trained and/or adjusted before it can be considered ready for use.

Front

What is the purpose of training and testing a machine learning model?

Back

It should be able to generalize new, unseen data.

Answer explanation

"It should be able to generalize new, unseen data" is the correct goal of training and testing a machine learning model.

"It will work perfectly on the corpus, but no new images can be added" is not true, as the purpose is to train a model that can generalize to new, unseen data.

"It will categorize images more accurately than a human" is not necessarily true, as the accuracy of the model depends on the quality of the dataset, the complexity of the problem, and the performance metrics used to evaluate the model.

"Its purpose is to be very good at distinguishing between different types of objects, such as different kinds of fish, but it won’t be very good at recognizing that all of those objects have fish traits in common" is a misunderstanding of the purpose of machine learning. The goal is not to classify objects based on their common traits, but to create a model that can accurately generalize to new data.

Front

How confident do you feel about this topic?

Back

Very confident, Mostly confident, Somewhat confident, Not confident at all