Exam AWS Certified Machine Learning - Specialty topic 1 question 163 discussion

B, and E (frequency encoding)

BD？ any thought?

These methods ensure that the color data is represented numerically while preserving the information’s integrity and relevance for the regression model

Using frequency encoding may help in some contexts but can introduce bias, especially if the frequency of a color is not related to the rental rate. This method does not leverage the actual differences between colors.

In this scenario, the specialist should use one-hot encoding and RGB encoding to allow the regression model to learn from the Wall_Color data. One-hot encoding is a technique used to convert categorical data into numerical data. It creates new columns that store one-hot representation of colors. For example, a variable named color has three categories: red, green, and blue. After one-hot encoding RGB encoding can capture the intensity and hue of a color, but it may also introduce correlation among the three columns. Therefore, using both one-hot encoding and RGB encoding can providemore information to the regression model than using either one alone.

Here we have a non-ordinal categorical variable to receive a numerical conversion for the model. Letter A is wrong as it is not an ordinal variable. Letter C is wrong as we are not going to retain any significant information for the model. The best solutions would be: Letter B and E. Letter D would be very interesting, but it would generate a problem of information fragmentation: most models consider the variables as being independent of each other, and these 3 columns by definition would not be independent.

B, and E (frequency encoding)

A+B make sense to me

B for sure D. This approach involves breaking down each color into its Red, Green, and Blue components and creating separate columns for each component. This allows the model to capture the information about the intensity of each color component, which can be useful in predicting the target variable. A, C, and E are not suitable for encoding color data in a way that can be used by a regression model. The integer transformation approach in option A arbitrarily assigns values to colors without any meaningful relationship between them. The approach in option C replaces the color names with their length, which does not provide any useful information for the model. Option E replaces each color name with its frequency in the training set, which does not capture any information about the color itself.

I think frequency encoding cannot be. What if some colors have same amount of frequency?

B. Add new columns that store one-hot representation of colors. One-hot encoding is a common approach to represent categorical variables as numerical values. This approach creates new binary variables for each category and assigns a value of 1 to the corresponding category and 0 to the others. In this case, the specialist can create three new binary variables, one for each color (Red, White, and Green) and use them as input to the regression model. E. Replace each color name by its training set frequency. Another approach to convert categorical variables into numerical ones is to replace each category with its frequency of occurrence in the training set. In this case, the specialist can replace the color names with their respective frequencies (1/3 for Red, 1/3 for White, and 1/3 for Green) to represent them numerically.

These are the only options preserving what "color" is. One-hot encoding is a default standard for any categorical data to be fed to a model that takes in numeric input. RGB format is a good numeric representation of any color by preserving its nature

A&B https://victorzhou.com/blog/one-hot/#:~:text=One-Hot%20Encoding%20takes%20a%20single%20integer%20and%20produces,of%20colors%20are%20possible%3A%20red%2C%20green%2C%20or%20blue.

Frequency encoding

B and E

BE is correct. For e, please refer: https://medium.com/analytics-vidhya/different-type-of-feature-engineering-encoding-techniques-for-categorical-variable-encoding-214363a016fb#:~:text=One%20Hot%20Encoding%3A%20%E2%80%94%20In%20this,slows%20down%20the%20learning%20significantly.