Pipeline with GridSearchCV in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Pipeline with GridSearchCV

Which metric matters for Pipeline with GridSearchCV and WHY

When using a pipeline with GridSearchCV, the main goal is to find the best model settings that work well on new data. The metric you choose depends on your problem:

• Accuracy if classes are balanced and you want overall correctness.
• Precision if false alarms are costly (e.g., spam detection).
• Recall if missing positive cases is bad (e.g., disease detection).
• F1 score if you want a balance between precision and recall.

GridSearchCV uses this metric to compare different model setups inside the pipeline and pick the best one.

Confusion Matrix Example

Suppose after GridSearchCV finds the best model, you test it and get this confusion matrix:

      | Predicted Positive | Predicted Negative |
      |--------------------|--------------------|
      | True Positive (TP): 40 | False Negative (FN): 10 |
      | False Positive (FP): 5 | True Negative (TN): 45 |
    

Totals: 40 + 10 + 5 + 45 = 100 samples

From this, you calculate:

• Precision = TP / (TP + FP) = 40 / (40 + 5) = 0.89
• Recall = TP / (TP + FN) = 40 / (40 + 10) = 0.80
• Accuracy = (TP + TN) / Total = (40 + 45) / 100 = 0.85
• F1 Score = 2 * (Precision * Recall) / (Precision + Recall) ≈ 0.84

Precision vs Recall Tradeoff with Pipeline and GridSearchCV

GridSearchCV helps tune model settings to balance precision and recall. For example:

• If you want to catch all positive cases (high recall), you might accept more false alarms (lower precision).
• If you want to avoid false alarms (high precision), you might miss some positive cases (lower recall).

GridSearchCV tries many combinations to find the best balance based on your chosen metric.

Example:

• Spam filter: prioritize precision to avoid marking good emails as spam.
• Medical test: prioritize recall to catch all sick patients.

What "Good" vs "Bad" Metric Values Look Like

Good metrics depend on your problem and data, but here are general ideas:

• Good: Precision and recall both above 0.8, accuracy above 0.85, F1 score balanced and high.
• Bad: Very low precision (e.g., 0.3) means many false alarms.
• Very low recall (e.g., 0.2) means many missed positives.
• Accuracy can be misleading if classes are imbalanced.

GridSearchCV helps find settings that improve these metrics by testing many options.

Common Pitfalls When Using Pipeline with GridSearchCV

• Data leakage: Including test data in training or preprocessing before splitting can give overly optimistic results.
• Overfitting: GridSearchCV may pick a model that fits training data too well but fails on new data.
• Ignoring metric choice: Using accuracy on imbalanced data can hide poor performance on minority class.
• Not using cross-validation: Without proper splitting, results may not generalize.

Self Check

Your pipeline with GridSearchCV found a model with 98% accuracy but only 12% recall on fraud cases. Is this good for production?

Answer: No. Even though accuracy is high, the model misses 88% of fraud cases (low recall). For fraud detection, catching fraud (high recall) is critical. This model would let most fraud slip through.

Key Result

GridSearchCV uses your chosen metric (like precision, recall, or F1) to find the best model settings, balancing tradeoffs to improve real-world performance.

Practice

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

easy

A. To combine preprocessing steps and model training into one object

B. To speed up the training by using multiple CPUs

C. To automatically select the best model type

D. To visualize the model's decision boundaries

Solution

1. Step 1: Understand what a Pipeline does

   A Pipeline chains preprocessing and model training steps so they run together smoothly.

2. Step 2: Identify the main benefit

   This chaining helps avoid mistakes and makes code cleaner by combining steps into one object.

3. Final Answer:

   To combine preprocessing steps and model training into one object -> Option A

4. Quick Check:

   Pipeline = combine steps [OK]

Hint: Pipeline bundles steps to simplify workflow [OK]

Common Mistakes:

• Thinking Pipeline speeds up training automatically
• Confusing Pipeline with model selection
• Believing Pipeline creates visualizations

2. Which syntax correctly sets the parameter n_estimators of a RandomForest inside a pipeline named pipe for GridSearchCV?

easy

A. {'randomforest-n_estimators': [10, 50, 100]}

B. {'random_forest__n_estimators': [10, 50, 100]}

C. {'randomforest.n_estimators': [10, 50, 100]}

D. {'randomforest__n_estimators': [10, 50, 100]}

Solution

1. Step 1: Recall parameter naming in Pipeline

   Parameters inside a pipeline step use double underscores: stepname__paramname.

2. Step 2: Match step name and parameter

   If the step is named 'randomforest', then 'randomforest__n_estimators' is correct syntax.

3. Final Answer:

   {'randomforest__n_estimators': [10, 50, 100]} -> Option D

4. Quick Check:

   Use double underscores between step and param [OK]

Hint: Use double underscores between step and parameter [OK]

Common Mistakes:

• Using single underscore instead of double
• Using dot or dash instead of double underscore
• Misspelling the pipeline step name

3. Given the code below, what will grid.best_params_ output?

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('clf', RandomForestClassifier(random_state=42))
])

param_grid = {'clf__n_estimators': [20], 'clf__max_depth': [4]}
grid = GridSearchCV(pipe, param_grid, cv=2)
grid.fit(X_train, y_train)

print(grid.best_params_)

medium

A. SyntaxError due to param_grid keys

B. {'clf__n_estimators': 10, 'clf__max_depth': 2}

C. {'clf__n_estimators': 20, 'clf__max_depth': 4}

D. KeyError because 'clf' is not a pipeline step

Solution

1. Step 1: Understand pipeline and param_grid

   The pipeline has a step named 'clf' for RandomForestClassifier. The param_grid uses 'clf__' prefix correctly.

2. Step 2: Determine the output

   Since param_grid specifies only one combination, GridSearchCV will select {'clf__n_estimators': 20, 'clf__max_depth': 4} as the best parameters.

3. Final Answer:

   {'clf__n_estimators': 20, 'clf__max_depth': 4} -> Option C

4. Quick Check:

   Best params match the only tested values [OK]

Hint: With single param values, they become best_params_ [OK]

Common Mistakes:

• Confusing step name 'clf' with 'classifier'
• Using single underscore in param_grid keys
• Assuming syntax error without checking keys

4. Identify the error in this pipeline and GridSearchCV setup:

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('model', RandomForestClassifier())
])

param_grid = {'randomforest__n_estimators': [10, 50]}
grid = GridSearchCV(pipe, param_grid)
grid.fit(X_train, y_train)

medium

A. The param_grid key should be 'model__n_estimators', not 'randomforest__n_estimators'

B. RandomForestClassifier cannot be used inside a pipeline

C. StandardScaler should not be the first step

D. GridSearchCV requires cv parameter

Solution

1. Step 1: Check pipeline step names

   The pipeline step for RandomForestClassifier is named 'model', not 'randomforest'.

2. Step 2: Match param_grid keys to pipeline steps

   Parameter keys must use the step name 'model' with double underscores, so 'model__n_estimators' is correct.

3. Final Answer:

   The param_grid key should be 'model__n_estimators', not 'randomforest__n_estimators' -> Option A

4. Quick Check:

   Param keys must match pipeline step names [OK]

Hint: Param keys must match pipeline step names exactly [OK]

Common Mistakes:

• Using wrong step name in param_grid keys
• Thinking RandomForest can't be in pipeline
• Believing cv is mandatory (it defaults to 5)

5. You want to tune both a scaler and a classifier in a pipeline using GridSearchCV. Which param_grid correctly tests StandardScaler with and without scaling, and RandomForest with 10 or 50 trees?

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('clf', RandomForestClassifier(random_state=0))
])

param_grid = ?

hard

A. {'scaler__': [StandardScaler(), None], 'clf__n_estimators': [10, 50]}

B. {'scaler': [StandardScaler(), None], 'clf__n_estimators': [10, 50]}

C. {'scaler': [StandardScaler(), None], 'clf__n_estimators': [10, 50], 'clf__max_depth': [None]}

D. {'scaler__with_mean': [True, False], 'clf__n_estimators': [10, 50]}

Solution

1. Step 1: Understand how to toggle scaler on/off in pipeline

   To test with and without scaling, replace the scaler step with StandardScaler() or None in param_grid using the step name 'scaler'.

2. Step 2: Set classifier parameters correctly

   Use 'clf__n_estimators' to test 10 and 50 trees for the RandomForestClassifier step named 'clf'.

3. Final Answer:

   {'scaler': [StandardScaler(), None], 'clf__n_estimators': [10, 50]} -> Option B

4. Quick Check:

   Toggle scaler by replacing step, tune clf params with double underscores [OK]

Hint: Toggle steps by replacing with None in param_grid [OK]

Common Mistakes:

• Trying to set scaler params with double underscores incorrectly
• Using 'scaler__' key with no param name
• Not using None to disable a step