RandomizedSearchCV in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - RandomizedSearchCV

Which metric matters for RandomizedSearchCV and WHY

RandomizedSearchCV helps find the best model settings by trying many options randomly. The metric you choose to guide this search depends on your goal. For example, if you want to catch all positive cases, use Recall. If you want to be sure predictions are correct, use Precision. For balanced performance, use F1 score. The metric tells RandomizedSearchCV which model settings are best.

Confusion matrix example

      Actual \ Predicted | Positive | Negative
      -------------------|----------|---------
      Positive           |    80    |   20    
      Negative           |    10    |   90    

      Total samples = 200

      Precision = TP / (TP + FP) = 80 / (80 + 10) = 0.89
      Recall = TP / (TP + FN) = 80 / (80 + 20) = 0.80
      F1 = 2 * (0.89 * 0.80) / (0.89 + 0.80) = 0.84
    

Precision vs Recall tradeoff with examples

RandomizedSearchCV can optimize for different metrics, but improving one can reduce the other.

• Spam filter: High Precision is important to avoid marking good emails as spam.
• Cancer detection: High Recall is critical to catch all cancer cases, even if some false alarms happen.

Choosing the right metric guides RandomizedSearchCV to find model settings that balance these needs.

Good vs Bad metric values for RandomizedSearchCV tuning

Good: High F1 score (close to 1), balanced Precision and Recall, consistent results on new data.

Bad: High accuracy but very low Recall or Precision, large difference between training and test scores (overfitting), unstable results.

Common pitfalls when using metrics with RandomizedSearchCV

• Accuracy paradox: High accuracy can be misleading if classes are imbalanced.
• Data leakage: Using test data during tuning inflates metric scores falsely.
• Overfitting: Very high training scores but low test scores mean the model memorizes data, not learns patterns.
• Ignoring metric choice: Using default accuracy when Recall or Precision matters can lead to poor real-world results.

Self-check question

Your model tuned with RandomizedSearchCV 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. The low Recall means the model misses most fraud cases, which is dangerous. Despite high accuracy, it fails the main goal of catching fraud.

Key Result

RandomizedSearchCV uses your chosen metric (like Precision, Recall, or F1) to find the best model settings balancing your real-world needs.

Practice

1. What is the main purpose of using RandomizedSearchCV in machine learning?

easy

A. To find the best model parameters by trying random combinations

B. To train the model on the entire dataset without validation

C. To reduce the size of the dataset by random sampling

D. To visualize the model's predictions on random data points

Solution

1. Step 1: Understand the role of RandomizedSearchCV

   RandomizedSearchCV tries random combinations of model parameters to find the best settings efficiently.

2. Step 2: Compare with other options

   Options B, C, and D describe unrelated tasks like training without validation, data sampling, or visualization, which are not the purpose of RandomizedSearchCV.

3. Final Answer:

   To find the best model parameters by trying random combinations -> Option A

4. Quick Check:

   RandomizedSearchCV = random parameter search [OK]

Hint: RandomizedSearchCV tries random settings to find best model [OK]

Common Mistakes:

• Confusing RandomizedSearchCV with data sampling methods
• Thinking it trains without validation
• Assuming it visualizes data

2. Which of the following is the correct way to import RandomizedSearchCV from scikit-learn?

easy

A. import RandomizedSearchCV from sklearn.model_selection

B. from sklearn.random import RandomizedSearchCV

C. from sklearn.model_selection import RandomizedSearchCV

D. from sklearn.search import RandomizedSearchCV

Solution

1. Step 1: Recall the correct import path

   RandomizedSearchCV is part of the model_selection module in scikit-learn, so the correct import is from sklearn.model_selection.

2. Step 2: Check syntax correctness

   from sklearn.model_selection import RandomizedSearchCV uses the correct Python import syntax. Options B and D use wrong module names, and import RandomizedSearchCV from sklearn.model_selection uses incorrect import syntax.

3. Final Answer:

   from sklearn.model_selection import RandomizedSearchCV -> Option C

4. Quick Check:

   Correct import = sklearn.model_selection [OK]

Hint: Remember RandomizedSearchCV is in sklearn.model_selection [OK]

Common Mistakes:

• Using wrong module names like sklearn.random
• Incorrect import syntax like 'import X from Y'
• Confusing with other sklearn modules

3. Given the code below, what will print(random_search.best_params_) output?

from sklearn.datasets import load_iris
from sklearn.svm import SVC
from sklearn.model_selection import RandomizedSearchCV
import numpy as np

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

param_dist = {'C': [0.1, 1, 10], 'kernel': ['linear', 'rbf']}
model = SVC()
random_search = RandomizedSearchCV(model, param_distributions=param_dist, n_iter=2, random_state=42)
random_search.fit(X, y)
print(random_search.best_params_)

medium

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

B. Raises an error because n_iter is too small

C. {'C': 0.1, 'kernel': 'linear'}

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

Solution

1. Step 1: Understand RandomizedSearchCV behavior with n_iter=2

   RandomizedSearchCV tries 2 random parameter combinations from the given param_dist. With random_state=42, the selected combos are reproducible.

2. Step 2: Check which parameters are chosen

   Running this code selects {'C': 10, 'kernel': 'rbf'} as best params based on scoring on iris data.

3. Final Answer:

   {'C': 10, 'kernel': 'rbf'} -> Option A

4. Quick Check:

   Random seed + n_iter=2 picks {'C':10,'kernel':'rbf'} [OK]

Hint: RandomizedSearchCV picks random combos; seed fixes output [OK]

Common Mistakes:

• Assuming all parameter combos are tried
• Thinking n_iter too small causes error
• Ignoring random_state effect on results

4. You wrote this code but get an error: TypeError: RandomizedSearchCV.__init__() got an unexpected keyword argument 'param_grid'. What is the mistake?

from sklearn.model_selection import RandomizedSearchCV
from sklearn.tree import DecisionTreeClassifier
param_grid = {'max_depth': [3, 5, 7]}
random_search = RandomizedSearchCV(estimator=DecisionTreeClassifier(), param_grid=param_grid, n_iter=3)
random_search.fit(X_train, y_train)

medium

A. n_iter must be larger than the number of parameters

B. You must import DecisionTreeClassifier before using it

C. RandomizedSearchCV does not accept n_iter parameter

D. Use 'param_distributions' instead of 'param_grid' for RandomizedSearchCV

Solution

1. Step 1: Identify the error cause

   The error says 'param_grid' is unexpected. RandomizedSearchCV expects 'param_distributions' instead of 'param_grid'.

2. Step 2: Correct the parameter name

   Replacing 'param_grid' with 'param_distributions' fixes the error and allows random search to work.

3. Final Answer:

   Use 'param_distributions' instead of 'param_grid' for RandomizedSearchCV -> Option D

4. Quick Check:

   RandomizedSearchCV needs param_distributions [OK]

Hint: RandomizedSearchCV uses param_distributions, not param_grid [OK]

Common Mistakes:

• Confusing RandomizedSearchCV with GridSearchCV parameter names
• Ignoring error message details
• Assuming n_iter is invalid parameter

5. You want to tune a Random Forest model with many parameters but have limited time. Which approach best uses RandomizedSearchCV to find good parameters efficiently?

hard

A. Set n_iter equal to all possible parameter combinations for exhaustive search

B. Set a reasonable n_iter to try a random subset of parameter combos and use cross-validation

C. Use RandomizedSearchCV without specifying param_distributions to try defaults

D. Use RandomizedSearchCV with n_iter=1 to save time

Solution

1. Step 1: Understand the goal of RandomizedSearchCV

   It helps find good parameters by trying a random sample of combinations, which is faster than exhaustive search.

2. Step 2: Choose the best strategy for limited time

   Setting n_iter to a reasonable number balances search quality and time. Using cross-validation ensures reliable results.

3. Final Answer:

   Set a reasonable n_iter to try a random subset of parameter combos and use cross-validation -> Option B

4. Quick Check:

   RandomizedSearchCV = random subset + cross-validation [OK]

Hint: Balance n_iter and validation for efficient tuning [OK]

Common Mistakes:

• Trying all combos defeats RandomizedSearchCV purpose
• Not specifying param_distributions causes errors
• Using n_iter=1 usually gives poor results