NCA-GENM Question 58
Select 3A data scientist is working on a multimodal generative AI project that involves text and image data. During the data preparation phase, they notice that the text data contains missing values, duplicate entries, and inconsistent formatting, while the image data has varying resolutions and color formats. Which of the following actions should they prioritize to ensure the data is ready for modeling?
- A
Remove duplicate entries and standardize text formatting in the text data.
- B
Impute missing values in the text data using appropriate statistical or ML-based methods.
- C
Resize all images to a consistent resolution and convert their color format to a uniform standard.
- D
Directly train the model without addressing the inconsistencies in the text and image data.
- E
Generate synthetic data to compensate for missing values and improve model performance.
Show answer and explanation
Correct answers: A, B, C
Explanation
Inspecting, cleansing, and transforming data are critical steps in preparing multimodal datasets for generative AI models. Addressing issues such as duplicates, missing values, and inconsistencies ensures the data is reliable and suitable for training. For text data, cleaning involves deduplication, formatting standardization, and imputation of missing values, while for image data, resizing and format standardization are necessary for uniformity. Skipping these steps or relying solely on synthetic data could lead to suboptimal model performance.
- A. Correct.
Removing duplicates and standardizing text formatting is a critical step in cleaning the text data to ensure consistency and avoid introducing noise into the model.
- B. Correct.
Imputing missing values is essential to handle incomplete data and prevent loss of information, which could negatively impact the model’s performance.
- C. Correct.
Resizing images to a consistent resolution and standardizing their color format ensures uniformity, making the data suitable for training computer vision models.
- D. Incorrect.
Skipping data cleaning and directly training the model would likely result in poor model performance due to the presence of noise and inconsistencies in the data.
- E. Incorrect.
While synthetic data generation can be useful in certain scenarios, it is not an immediate solution for handling missing values or inconsistencies in the existing dataset.