Free Microsoft DP-100 Practice Test Questions MCQs

Explanation
The question asks for a metric to determine how closely the data fits the regression line. This refers to "goodness of fit," which measures the proportion of the variance in the dependent variable that is predictable from the independent variable(s). While several metrics can evaluate a regression model's performance, only one directly quantifies the proportion of variance explained by the model relative to the total variance.

Correct Option

A. Coefficient of determination
The Coefficient of determination, commonly denoted as R-squared (R²), is the statistical metric that specifically measures how close the data are to the fitted regression line.

It represents the proportion of the variance in the dependent variable that is predictable from the independent variable(s). An R² of 100% indicates that all changes in the dependent variable are completely explained by changes in the independent variable(s).

It is the direct answer to determining the "goodness of fit" for a regression model.

Incorrect Option

B. Recall
Recall is a classification metric, not a regression metric. It measures the proportion of actual positives that were correctly identified by the model (True Positives / (True Positives + False Negatives)).

It is used in scenarios like binary classification (e.g., spam detection, disease diagnosis) and is irrelevant for evaluating linear regression fits.

C. Precision
Precision is also a classification metric. It measures the proportion of positive identifications that were actually correct (True Positives / (True Positives + False Positives)).

Like recall, it is used to evaluate classification models and does not apply to the context of a regression line.

D. Mean absolute error
Mean Absolute Error (MAE) is a valid regression metric, but it measures the average magnitude of the errors in a set of predictions, without considering their direction. It calculates the average absolute difference between the predicted and actual values.

While it tells you how wrong the predictions are on average, it does not quantify how well the data fits the line (the proportion of variance explained).

E. Root Mean Square Error
Root Mean Square Error (RMSE) is another standard regression metric. It measures the average magnitude of the error by calculating the square root of the average of squared differences between prediction and actual observation.

RMSE gives a relatively high weight to large errors. However, like MAE, it measures prediction error magnitude, not the proportion of variance explained by the model fit.

Reference
Microsoft Learn: Train and evaluate regression models

Explanation
The question requires recording a simple numeric value (the row count) as a metric during an Azure Machine Learning experiment run. The metric must be retrievable after the run using the get_metrics() method. Azure ML provides different logging methods for different data types, and for a single numeric value, the appropriate method is the one that logs a key-value pair to the run's metrics record.

Correct Option

B. run.log('row_count', rows)
The log() function is specifically designed to record a single numeric value as a metric in an Azure ML run.

It creates a key-value pair where 'row_count' is the metric name, and the variable rows (containing the integer value) is the metric value.

After execution, this value can be retrieved using run.get_metrics()['row_count'] and will appear in the Azure ML studio under the run's metrics tab.

Incorrect Option

A. run.upload_file('row_count', './data.csv')
The upload_file() method is used to upload files (like models or datasets) to the run's output storage, not to log metrics.

This would upload the entire CSV file to a location named 'row_count', which is not the intended behavior for tracking a simple row count metric.

Uploaded files are accessible through the "Outputs + logs" section, not through the get_metrics() method.

C. run.tag('row_count', rows)
The tag() method is used to add metadata tags to the run itself for organization and searchability, not for logging experiment metrics.

Tags are typically used for filtering and grouping runs (e.g., 'experiment_type': 'regression'), not for tracking numerical results.

Tags cannot be retrieved using the get_metrics() method.

D. run.log_table('row_count', rows)
The log_table() method is used to log dictionary-like objects or tables where the data has multiple columns and rows.

The rows variable in this context is a single integer (the length of the DataFrame), not a table structure.

Using this for a scalar value would either fail or store it incorrectly as a table with unexpected format.

E. run.log_row('row_count', rows)
The log_row() method is used to log a metric row by row, typically when you want to create a table with multiple columns and append multiple rows over time.

It expects column names and values as key-value pairs (e.g., run.log_row("MyTable", column1=value1, column2=value2)).

Passing a single value rows without a column name specification would result in an error.

Reference
Microsoft Learn: Log metrics in Azure ML experiments

Explanation
The goal requires deploying a model as a real-time web service with key-based authentication. Key-based authentication in Azure Machine Learning endpoints uses either primary and secondary keys. The solution incorrectly configures authentication properties, which determines whether the deployed service will meet the requirement.

Correct Option

B. No
The solution does not meet the goal because it configures conflicting authentication settings.

Setting auth_enabled=False disables key-based authentication entirely, which directly contradicts the requirement for key-based authentication.

Setting token_auth_enabled=True enables token-based authentication (Azure Active Directory tokens), which is a different authentication method than key-based.

For key-based authentication, you must set auth_enabled=True (which is the default value) and ensure token authentication is disabled.

The correct configuration would be to leave auth_enabled as True (or explicitly set it to True) and set token_auth_enabled=False.

Reference
Microsoft Learn: Authentication for Azure ML web services

Explanation
The scenario requires loading all sales data to date while maintaining the ability to filter data by time. The folder structure uses wildcard-compatible patterns with month/year folders. The key requirements are loading all data easily, filtering by specific months, and registering the minimum number of datasets. Understanding Azure ML dataset path patterns and versioning strategies is essential to meet these requirements.

Correct Option

B. Create a tabular dataset that references the datastore and specifies the path 'sales/*/sales.csv', register the dataset with the name sales_dataset and a tag named month indicating the month and year it was registered, and use this dataset for all experiments.
The path pattern 'sales//sales.csv' uses a wildcard () to include all month-year subfolders, automatically incorporating new monthly data as it is added.

This satisfies loading all sales data to date with a single dataset registration, meeting the minimum dataset requirement.

Since the dataset loads all available data, filtering for experiments that require data before a specific month can be done in code after loading the dataframe, using the month tags from folder names or date columns in the data.

Registering with a month tag provides metadata about when the dataset was last updated, though it is not strictly necessary for the filtering requirement.

Incorrect Option

A. Create a tabular dataset that references the datastore and explicitly specifies each 'sales/mm-yyyy/sales.csv' file every month. Register the dataset with the name sales_dataset each month, replacing the existing dataset and specifying a tag named month indicating the month and year it was registered. Use this dataset for all experiments.
This approach requires manually updating the dataset definition every month to include the new file path.

Replacing the existing dataset each month loses the ability to reference previous dataset states, making it impossible to reproduce experiments that used earlier data without manual intervention.

This violates the requirement to use data from before a specific previous month, as the overwritten dataset only contains the latest configuration.

C. Create a new tabular dataset that references the datastore and explicitly specifies each 'sales/mm-yyyy/sales.csv' file every month. Register the dataset with the name sales_dataset_MM-YYYY each month with appropriate MM and YYYY values for the month and year. Use the appropriate month-specific dataset for experiments.
This creates a new dataset each month with a month-specific name, resulting in many registered datasets over time.

While this allows precise filtering by selecting the appropriate month-specific dataset, it violates the requirement to register the minimum number of datasets possible.

It also requires manual creation and management of multiple datasets instead of leveraging path patterns.

D. Create a tabular dataset that references the datastore and explicitly specifies each 'sales/mm-yyyy/sales.csv' file. Register the dataset with the name sales_dataset each month as a new version and with a tag named month indicating the month and year it was registered. Use this dataset for all experiments, identifying the version to be used based on the month tag as necessary.
This approach requires updating the dataset definition each month to add the new file path and creating a new version.

While versioning preserves history, it still requires manual updates and creates multiple versions, though they share the same name.

The requirement to filter by specific months would require selecting the correct version based on tags, but each version contains all files up to that point, not just a specific month, making month-specific filtering impossible without post-load filtering.

Reference
Microsoft Learn: Create Azure Machine Learning datasets

Explanation
The goal requires cleaning missing values without affecting the dimensionality of the feature set. Dimensionality refers to the number of features (columns) in the dataset. The solution must include all values in the final dataset. The proposed median imputation addresses missing values but fails to meet the requirement of analyzing a full dataset to include all values.

Correct Option

B. No
The solution does not meet the goal because the question specifies the need to "analyze a full dataset to include all values," which means examining the data before cleaning to understand the pattern of missing values.

Using median imputation without first analyzing the missing data ignores important information about why values are missing and their distribution.

A full analysis should include visualization of missing data patterns, calculating the percentage of missing values per column, and determining if missingness is random or systematic.

The solution jumps directly to imputation without performing the required analysis step, which violates the requirement to analyze the full dataset including missing values.

Reference
Microsoft Learn: Handle missing data in Azure Machine Learning

Explanation
Poisson regression is specifically designed for modeling count data, which has distinct mathematical requirements. The question asks for conditions that the feature set must contain, but actually focuses on the properties of the label (target) variable since Poisson regression assumptions primarily concern the dependent variable. Understanding these assumptions is critical for correctly applying this regression technique.

Correct Option

C. The label data must be a positive value
Poisson regression models count data, which by definition cannot be negative. Counts represent the number of occurrences of an event (e.g., number of calls, number of customers).

The Poisson distribution is defined only for non-negative integers, and the model uses a log link function that requires the mean to be positive.

If the label contains negative values, Poisson regression cannot be applied, and alternative models like linear regression or data transformation would be necessary.

E. The data must be whole numbers
Poisson regression requires the dependent variable to consist of non-negative integers (whole numbers: 0, 1, 2, 3, ...).

This is because the Poisson distribution models the probability of a given number of events occurring in a fixed interval, which is inherently discrete.

Continuous values or decimals would violate the fundamental assumption of the Poisson distribution and would require different modeling approaches such as Gaussian regression or Gamma regression.

Incorrect Option

A. The label data must be a negative value
This is completely incorrect as Poisson regression cannot handle negative values. The Poisson distribution is defined only for non-negative integers.

Negative counts are impossible in real-world count data scenarios like number of calls, accidents, or customers.

B. The label data can be positive or negative
This is incorrect because Poisson regression cannot accommodate negative values. If the label contains negative values, the model assumptions are violated.

While positive values are acceptable, the presence of any negative values makes Poisson regression invalid.

D. The label data must be non discrete
This is incorrect as Poisson regression specifically requires discrete data (counts). Non-discrete (continuous) data would violate the Poisson distribution assumption.

For continuous positive data, other regression techniques like Gamma regression or log-transformed linear regression would be more appropriate.

Reference
Microsoft Learn: Poisson regression in Azure Machine Learning

Explanation
In Azure Machine Learning Studio (classic) or Azure Machine Learning designer, dividing data into two distinct datasets is a common data preparation task, typically for creating training and testing sets. The question asks for the specific module that performs this operation. Understanding the function of each module helps in selecting the correct one.

Correct Option

A. Partition and Sample
The Partition and Sample module is specifically designed to divide a dataset into two or more distinct subsets based on various strategies.

It supports splitting data using techniques like "Split into partitions" where you can specify a fraction of data for the first partition (e.g., 0.7 for training) and the remaining for the second partition (e.g., 0.3 for testing).

It also allows stratified splitting to maintain class distributions, which is essential for imbalanced datasets.

This module directly fulfills the requirement to create two distinct datasets.

Incorrect Option

B. Assign Data to Clusters
The Assign Data to Clusters module is used after training a clustering model (like K-Means) to assign new data points to existing clusters based on their distances to cluster centroids.

It is not designed for splitting a single dataset into two distinct datasets but rather for classification or assignment tasks based on an already trained model.

C. Group Data into Bins
The Group Data into Bins module (also known as discretization) is used to convert continuous numerical data into categorical bins or intervals (e.g., grouping ages into 0-18, 19-35, etc.).

Its purpose is data transformation, not data splitting. It changes the values within a column rather than creating separate datasets.

D. Test Hypothesis Using t-Test
The Test Hypothesis Using t-Test module is a statistical analysis tool used to determine if there is a significant difference between the means of two groups. It is for analysis and inference, not for data preparation or splitting datasets into two parts.

Reference
Microsoft Learn: Partition and Sample module

Explanation
The question requires understanding how the model makes predictions and specifically checking if features like geographic location (where an applicant lives) influence predictions, which could violate regulations. This is a model interpretability and explainability problem. Azure Machine Learning provides tools to explain feature importance and model behavior to ensure fairness and regulatory compliance.

Correct Option

D. Use the interpretability package to generate an explainer for the model.
The Azure Machine Learning interpretability package provides tools to explain model predictions by calculating feature importance scores at both global and local levels.

By generating an explainer, you can determine the extent to which each feature (including location-based features) influences predictions, directly addressing the regulatory concern.

It supports various explanation techniques like SHAP (SHapley Additive exPlanations), Mimic Explainer, and Permutation Feature Importance.

The results can show whether protected attributes like geographic location are having an inappropriate impact on loan decisions.

Incorrect Option

A. Enable data drift monitoring for the model and its training dataset.
Data drift monitoring detects when the statistical properties of the input data change over time compared to the training data.

While important for maintaining model performance, it does not explain feature influence or help determine if specific features are improperly affecting predictions.

This addresses model performance degradation, not interpretability or fairness.

B. Score the model against some test data with known label values and use the results to calculate a confusion matrix.
Scoring and confusion matrices evaluate model performance metrics like accuracy, precision, recall, and F1 score.

These metrics measure how well the model predicts but provide no insight into which features drive those predictions.

This approach cannot identify whether location or other protected attributes influence decisions.

C. Use the Hyperdrive library to test the model with multiple hyperparameter values.
Hyperdrive is used for hyperparameter tuning to find the optimal model configuration that maximizes performance.

It optimizes model accuracy but does not provide explanations about feature importance or model behavior.

This addresses model optimization, not interpretability or regulatory compliance checking.

E. Add tags to the model registration indicating the names of the features in the training dataset.
Adding tags to model registration is useful for metadata management, organization, and tracking.

While documenting feature names is good practice, tags themselves do not provide any analysis of feature influence on predictions.

This is metadata storage, not model explanation.

Reference
Microsoft Learn: Model interpretability in Azure Machine Learning

Explanation
The goal requires creating a pipeline where a processing script loads data from a datastore and passes the processed data to a training script. The provided code attempts to define a two-step pipeline with process_step and train_step. The key requirement is correct data flow between steps, where the processed data from the first step becomes the input to the second step.

Correct Option

B. No
The solution fails because the data flow between steps is incorrectly configured. In the code, process_step takes data_input as an argument and produces data_output.

However, train_step incorrectly uses data_input as its data source instead of using the data_output from process_step. This means the training script receives unprocessed raw data directly from the datastore, not the processed data.

The inputs=[data_output] parameter in train_step specifies that data_output is an input, but the script arguments still reference data_input.

For the pipeline to work correctly, the train_step arguments should reference data_output instead of data_input, ensuring the processed data flows properly between steps.

Reference
Microsoft Learn: Build Azure Machine Learning pipelines

Explanation
Precision is a metric used to evaluate binary classification models. The question asks which visualization should be used when precision is the evaluation metric. However, precision itself is not a visualization but a calculated metric. Among the options, only one visualization is commonly associated with binary classification evaluation and can help understand the trade-off between precision and other metrics.

Correct Option

C. Receiver Operating Characteristic (ROC) curve
The ROC curve is a fundamental visualization tool for binary classification evaluation that plots the True Positive Rate (Sensitivity) against the False Positive Rate (1-Specificity) at various threshold settings.

While the ROC curve does not directly display precision, it is closely related to the overall classification performance. The Area Under the Curve (AUC) derived from the ROC curve is a summary metric of model performance.

Precision can be calculated from the confusion matrix, and there are other curves like Precision-Recall curves that are more directly related to precision, especially for imbalanced datasets.

Among the given options, the ROC curve is the only visualization that is standard for binary classification evaluation.

Incorrect Option

A. scatter plot
Scatter plots are used to visualize the relationship between two continuous variables, showing individual data points as dots in a Cartesian coordinate system.

They are not used for evaluating classification model performance or visualizing metrics like precision.

B. coefficient of determination
The coefficient of determination (R²) is a metric used for regression models, not classification models. It measures the proportion of variance in the dependent variable explained by the independent variables.

This is not a visualization and is irrelevant for binary classification evaluation.

D. Gradient descent
Gradient descent is an optimization algorithm used to minimize the loss function during model training by iteratively updating model parameters.

It is not a visualization or evaluation metric for completed models, but rather a training technique.

Reference
Microsoft Learn: Evaluate binary classification models