Hyperparameter tuning (GridSearchCV) in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Hyperparameter tuning (GridSearchCV)

Which metric matters for Hyperparameter tuning (GridSearchCV) and WHY

When tuning hyperparameters with GridSearchCV, the metric you choose depends on your goal. For example, if you want to balance correct positive and negative predictions, accuracy might be used. But if your problem needs catching all positives (like disease detection), recall is better. GridSearchCV uses this metric to pick the best settings that improve your model's performance on unseen data.

Confusion matrix example

      Actual \ Predicted | Positive | Negative
      -------------------|----------|---------
      Positive           |    TP=40 |   FN=10
      Negative           |    FP=5  |   TN=45
    

This matrix shows how many samples were correctly or wrongly predicted. GridSearchCV tries to find hyperparameters that improve these numbers according to the chosen metric.

Precision vs Recall tradeoff with examples

Precision means how many predicted positives are actually positive. Recall means how many actual positives were found.

Example 1: Spam filter - high precision is important to avoid marking good emails as spam.

Example 2: Cancer detection - high recall is important to catch all cancer cases, even if some false alarms happen.

GridSearchCV can optimize for either metric depending on what matters more in your problem.

What good vs bad metric values look like for Hyperparameter tuning

Good: The chosen metric (accuracy, precision, recall, F1) improves on validation data after tuning. For example, recall increases from 0.6 to 0.85, showing better detection.

Bad: Metrics stay the same or get worse, or the model overfits (great training score but poor validation score).

Common pitfalls in metrics during Hyperparameter tuning

• Using accuracy alone on imbalanced data can be misleading.
• Data leakage: tuning on test data leads to over-optimistic metrics.
• Overfitting: very high training metric but low validation metric.
• Choosing the wrong metric for your problem goal.

Self-check question

Your model tuned with GridSearchCV has 98% accuracy but only 12% recall on fraud cases. Is it good for production? Why or why not?

Answer: No, it is not good. Although accuracy is high, the model misses most fraud cases (low recall). For fraud detection, catching fraud (high recall) is critical, so this model would fail in real use.

Key Result

Choosing the right metric in GridSearchCV is key to tuning hyperparameters that truly improve model performance for your specific goal.

Practice

1.

What is the main purpose of using GridSearchCV in machine learning?

easy

A. To split data into training and testing sets

B. To find the best combination of hyperparameters automatically

C. To visualize the model's predictions

D. To reduce the size of the dataset

Solution

1. Step 1: Understand what GridSearchCV does

   GridSearchCV tries different hyperparameter combinations to find the best settings for a model.

2. Step 2: Compare options with GridSearchCV's purpose

   Only To find the best combination of hyperparameters automatically describes this automatic search for best hyperparameters.

3. Final Answer:

   To find the best combination of hyperparameters automatically -> Option B

4. Quick Check:

   GridSearchCV = best hyperparameters [OK]

Hint: GridSearchCV = automatic hyperparameter search [OK]

Common Mistakes:

• Confusing GridSearchCV with data splitting
• Thinking it visualizes predictions
• Assuming it reduces dataset size

2.

Which of the following is the correct way to import GridSearchCV from sklearn?

easy

A. import GridSearchCV from sklearn

B. from sklearn.grid_search import GridSearchCV

C. from sklearn.model_selection import GridSearchCV

D. from sklearn.tuning import GridSearchCV

Solution

1. Step 1: Recall the correct sklearn module for GridSearchCV

   GridSearchCV is in the model_selection module of sklearn.

2. Step 2: Match the correct import syntax

   from sklearn.model_selection import GridSearchCV uses the correct module and syntax for import.

3. Final Answer:

   from sklearn.model_selection import GridSearchCV -> Option C

4. Quick Check:

   Correct import = from sklearn.model_selection import GridSearchCV [OK]

Hint: GridSearchCV is in sklearn.model_selection [OK]

Common Mistakes:

• Using deprecated sklearn.grid_search module
• Wrong import syntax like 'import GridSearchCV from sklearn'
• Assuming a nonexistent sklearn.tuning module

3.

What will be the output of the following code snippet?

from sklearn.datasets import load_iris
from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV

iris = load_iris()
X, y = iris.data, iris.target

param_grid = {'kernel': ['linear', 'rbf'], 'C': [1, 10]}
svc = SVC()
grid = GridSearchCV(svc, param_grid, cv=3)
grid.fit(X, y)
print(grid.best_params_)

medium

A. A dictionary with best kernel and C values found by cross-validation

B. {'kernel': 'rbf', 'C': 10}

C. {'kernel': 'linear', 'C': 1}

D. An error because 'cv' parameter is missing

Solution

1. Step 1: Understand GridSearchCV output

   GridSearchCV's best_params_ attribute returns the best hyperparameters found after testing all combinations.

2. Step 2: Analyze the code snippet

   The code runs GridSearchCV with kernel and C values, then prints the best combination as a dictionary.

3. Final Answer:

   A dictionary with best kernel and C values found by cross-validation -> Option A

4. Quick Check:

   grid.best_params_ = best hyperparameters dict [OK]

Hint: grid.best_params_ shows best hyperparameters found [OK]

Common Mistakes:

• Expecting fixed values instead of dynamic best params
• Thinking cv parameter is missing (it is present)
• Confusing output with training scores

4.

Identify the error in this GridSearchCV usage:

from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV

param_grid = {'C': [0.1, 1, 10]}
svc = SVC()
grid = GridSearchCV(svc, param_grid)
grid.fit(X_train, y_train)
print(grid.best_score_)

Assuming X_train and y_train are defined.

medium

A. Missing cross-validation parameter cv in GridSearchCV

B. Parameter grid keys should be strings with double underscores

C. SVC requires kernel parameter in param_grid

D. No error; code runs correctly and prints best score

Solution

1. Step 1: Check default behavior of GridSearchCV

   If cv is not given, GridSearchCV uses default 5-fold cross-validation.

2. Step 2: Verify param_grid and SVC usage

   Parameter grid keys like 'C' are correct; kernel is optional with default 'rbf'.

3. Final Answer:

   No error; code runs correctly and prints best score -> Option D

4. Quick Check:

   Default cv=5, param_grid keys correct [OK]

Hint: GridSearchCV uses 5-fold CV by default if cv missing [OK]

Common Mistakes:

• Assuming cv parameter is mandatory
• Thinking param_grid keys need double underscores
• Believing kernel must be in param_grid

5.

You want to tune a RandomForestClassifier using GridSearchCV. You want to try n_estimators values 50, 100 and max_depth values 5, 10. Which code snippet correctly sets up the grid search?

hard

A. param_grid = {'n_estimators': [50, 100], 'max_depth': [5, 10]}
rf = RandomForestClassifier()
grid = GridSearchCV(rf, param_grid, cv=4)

B. param_grid = {'RandomForestClassifier__n_estimators': [50, 100], 'RandomForestClassifier__max_depth': [5, 10]}
rf = RandomForestClassifier()
grid = GridSearchCV(rf, param_grid)

C. param_grid = {'n_estimators': 50 or 100, 'max_depth': 5 or 10}
rf = RandomForestClassifier()
grid = GridSearchCV(rf, param_grid)

D. param_grid = {'n_estimators': [50, 100], 'max_depth': [5, 10]}
rf = RandomForestClassifier()
grid = GridSearchCV(rf, param_grid, cv='invalid')

Solution

1. Step 1: Define param_grid correctly for RandomForestClassifier

   Use keys as hyperparameter names with list of values: {'n_estimators': [50, 100], 'max_depth': [5, 10]}.

2. Step 2: Setup GridSearchCV with valid cv parameter

   The correct snippet initializes GridSearchCV with the proper param_grid using lists and a valid integer cv value (e.g., cv=4). Distractors include invalid cv values like strings, incorrect parameter key prefixes, and non-list value syntax like 'or'.

3. Final Answer:

   param_grid = {'n_estimators': [50, 100], 'max_depth': [5, 10]} rf = RandomForestClassifier() grid = GridSearchCV(rf, param_grid, cv=4) -> Option A

4. Quick Check:

   Correct param_grid and valid cv = param_grid = {'n_estimators': [50, 100], 'max_depth': [5, 10]} rf = RandomForestClassifier() grid = GridSearchCV(rf, param_grid, cv=4) [OK]

Hint: Use param names as keys and valid cv number [OK]

Common Mistakes:

• Adding model name prefix to param_grid keys
• Using 'or' instead of list for values
• Setting invalid cv values like strings