Precision-recall curves in TensorFlow - Model Metrics & Evaluation

Metrics & Evaluation - Precision-recall curves

Which metric matters for Precision-Recall Curves and WHY

Precision-Recall curves show how well a model balances precision and recall at different thresholds. This is important when classes are imbalanced or when missing positive cases is costly. Precision tells us how many predicted positives are actually correct. Recall tells us how many real positives the model found. The curve helps us pick a threshold that fits our needs.

Confusion Matrix Example

      Actual Positive | Actual Negative
    -------------------------------------
    Predicted Positive | TP = 80    | FP = 20
    Predicted Negative | FN = 10    | TN = 90
    -------------------------------------
    Total samples = 200
    

From this matrix:

• Precision = 80 / (80 + 20) = 0.8
• Recall = 80 / (80 + 10) = 0.8889

Precision vs Recall Tradeoff with Examples

Imagine a spam filter:

• High precision: Few good emails marked as spam (low false alarms).
• High recall: Most spam emails caught (few missed spam).

If you want to avoid losing good emails, choose a threshold for high precision. If you want to catch all spam, choose high recall.

Precision-Recall curves help find this balance by showing precision and recall at many thresholds.

Good vs Bad Metric Values for Precision-Recall Curves

Good: A curve that stays near the top-right corner means both precision and recall are high across thresholds. The area under the curve (AUC-PR) close to 1.0 is excellent.

Bad: A curve near the bottom or diagonal means poor precision or recall. AUC-PR near 0.5 means the model is guessing randomly.

Common Pitfalls with Precision-Recall Metrics

• Ignoring class imbalance: Accuracy can be misleading; precision-recall curves focus on positives.
• Data leakage: Inflates precision and recall falsely.
• Overfitting: Very high precision and recall on training but poor on new data.
• Misinterpreting precision and recall: Precision is about correctness of positive predictions; recall is about coverage of actual positives.

Self Check

Your model has 98% accuracy but 12% recall on fraud cases. Is it good for production?

Answer: No. The model misses 88% of fraud cases (low recall), which is dangerous. High accuracy is misleading because fraud is rare. Focus on improving recall to catch more fraud.

Key Result

Precision-Recall curves help choose the best balance between precision and recall, especially for imbalanced data or costly errors.

Practice

1. What does a precision-recall curve primarily show in model evaluation?

easy

A. The loss value during training

B. The trade-off between precision and recall for different thresholds

C. The accuracy of the model on the test set

D. The number of false negatives only

Solution

1. Step 1: Understand precision and recall

   Precision measures correct positive predictions, recall measures how many actual positives are found.

2. Step 2: Understand what the curve shows

   The precision-recall curve plots precision against recall at different classification thresholds, showing their trade-off.

3. Final Answer:

   The trade-off between precision and recall for different thresholds -> Option B

4. Quick Check:

   Precision-recall curve = trade-off between precision and recall [OK]

Hint: Precision-recall curves show precision vs recall trade-off [OK]

Common Mistakes:

• Confusing precision-recall curve with accuracy plot
• Thinking it shows loss values
• Assuming it only shows false negatives

2. Which TensorFlow function is used to compute precision and recall values for plotting a precision-recall curve?

easy

A. tf.keras.metrics.Precision() and tf.keras.metrics.Recall()

B. tf.keras.losses.MeanSquaredError()

C. tf.data.Dataset.batch()

D. tf.random.uniform()

Solution

1. Step 1: Identify functions related to precision and recall

   TensorFlow provides tf.keras.metrics.Precision() and tf.keras.metrics.Recall() to compute these metrics.

2. Step 2: Exclude unrelated functions

   MeanSquaredError is a loss, Dataset.batch is for data batching, random.uniform generates random numbers.

3. Final Answer:

   tf.keras.metrics.Precision() and tf.keras.metrics.Recall() -> Option A

4. Quick Check:

   Precision and recall functions = tf.keras.metrics.Precision and tf.keras.metrics.Recall [OK]

Hint: Use tf.keras.metrics.Precision and tf.keras.metrics.Recall for these metrics [OK]

Common Mistakes:

• Using loss functions instead of metric functions
• Confusing data pipeline functions with metrics
• Using random functions for metrics

3. Given the following TensorFlow code snippet, what will be the shape of the precision and recall tensors?

import tensorflow as tf
from sklearn.metrics import precision_recall_curve
y_true = tf.constant([0, 1, 1, 0, 1])
y_scores = tf.constant([0.1, 0.4, 0.35, 0.8, 0.7])
precision, recall, thresholds = precision_recall_curve(y_true, y_scores)
print(precision.shape, recall.shape)

medium

A. (1,) (1,)

B. (5,) (5,)

C. (N,) (N,) where N is number of thresholds + 1

D. Scalar values

Solution

1. Step 1: Understand precision_recall_curve output

   The function returns precision and recall arrays for each threshold plus one extra point, so length is number of thresholds + 1.

2. Step 2: Analyze shapes

   Since thresholds vary, precision and recall are vectors, not scalars or fixed length equal to input size.

3. Final Answer:

   (N,) (N,) where N is number of thresholds + 1 -> Option C

4. Quick Check:

   Precision and recall shapes = vectors with length thresholds+1 [OK]

Hint: precision_recall_curve returns arrays, not scalars [OK]

Common Mistakes:

• Assuming output shapes match input size
• Expecting scalar precision and recall
• Confusing thresholds count with input length

4. The following code attempts to plot a precision-recall curve but raises an error. What is the likely cause?

import tensorflow as tf
import matplotlib.pyplot as plt

y_true = tf.constant([0, 1, 1, 0])
y_scores = tf.constant([0.2, 0.8, 0.6, 0.4])
precision, recall, thresholds = tf.metrics.PrecisionRecallCurve()(y_true, y_scores)
plt.plot(precision, recall)
plt.show()

medium

A. Precision and recall are swapped in the plot call

B. plt.plot requires thresholds as x-axis

C. y_true and y_scores must be numpy arrays, not tensors

D. tf.metrics.PrecisionRecallCurve() does not exist

Solution

1. Step 1: Check TensorFlow API correctness

   TensorFlow does not have tf.metrics.PrecisionRecallCurve(); typically use from sklearn.metrics import precision_recall_curve.

2. Step 2: Understand error cause

   Using a non-existent function causes an AttributeError before plotting.

3. Final Answer:

   tf.metrics.PrecisionRecallCurve() does not exist -> Option D

4. Quick Check:

   Correct approach is sklearn.metrics.precision_recall_curve [OK]

Hint: Use sklearn.metrics.precision_recall_curve, not tf.metrics [OK]

Common Mistakes:

• Using non-existent TensorFlow function for PR curve
• Swapping precision and recall axes
• Assuming tensors can't be plotted

5. You have a highly imbalanced dataset with 1% positive class. You want to evaluate your TensorFlow model's ability to detect positives. Which approach is best to use and why? A) Use accuracy metric only B) Use confusion matrix without thresholds C) Use loss value during training D) Use precision-recall curve to evaluate model

hard

A. Use precision-recall curve to evaluate model

B. Use confusion matrix without thresholds

C. Use loss value during training

D. Use accuracy metric only

Solution

1. Step 1: Understand dataset imbalance impact

   With 1% positives, accuracy can be misleading because predicting all negatives gives high accuracy.

2. Step 2: Choose appropriate evaluation metric

   Precision-recall curves focus on positive class detection and show trade-offs, ideal for imbalanced data.

3. Final Answer:

   Use precision-recall curve to evaluate model -> Option A

4. Quick Check:

   Imbalanced data = use precision-recall curve [OK]

Hint: For imbalanced data, prefer precision-recall curves [OK]

Common Mistakes:

• Relying on accuracy for imbalanced data
• Ignoring threshold effects in confusion matrix
• Using loss value as evaluation metric