Retraining strategies in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding when to retrain a model

You have a machine learning model deployed for predicting house prices. Over time, the model's accuracy drops significantly. What is the most likely reason you need to retrain the model?

A The model was trained with too many features initially.

B The model's code has a syntax error causing wrong predictions.

C The data distribution has changed, so the model no longer fits new data well.

D The model's training data was too large, causing slow predictions.

💡 Hint

Think about what happens when the world changes but the model stays the same.

When the data changes over time (called data drift), the model's predictions become less accurate. Retraining with new data helps the model adapt.

❓ Predict Output

intermediate

Output of incremental retraining code

What will be the output of the following Python code snippet that simulates incremental retraining?

ML Python

class Model:
    def __init__(self):
        self.data_count = 0
    def train(self, new_data):
        self.data_count += len(new_data)
    def predict(self):
        return self.data_count

model = Model()
model.train([1, 2, 3])
model.train([4, 5])
print(model.predict())

A None

B 5

C 2

D 3

💡 Hint

Count how many items were added in total.

The model trains twice, first with 3 items, then with 2 items. The total count is 5, which is returned by predict.

❓ Model Choice

advanced

Choosing a retraining strategy for streaming data

You have a model that receives continuous streaming data. Which retraining strategy is best to keep the model updated without retraining from scratch every time?

A Incremental learning that updates the model with new data only.

B Batch retraining using all historical data every day.

C Training a new model from scratch every hour.

D Freezing the model and never retraining.

💡 Hint

Think about efficiency and keeping the model fresh with new data.

Incremental learning updates the model with new data without full retraining, making it efficient for streaming data.

❓ Metrics

advanced

Evaluating retraining effectiveness

After retraining a classification model, which metric change best indicates the retraining improved the model?

A Accuracy increased and loss decreased on validation data.

B Training loss increased but validation accuracy increased.

C Validation loss increased and accuracy stayed the same.

D Accuracy decreased but training loss decreased.

💡 Hint

Look for improvements on data the model hasn't seen before.

Improved accuracy and lower loss on validation data show the model generalizes better after retraining.

🔧 Debug

expert

Debugging retraining code with data leakage

Consider this retraining code snippet. What is the main issue causing overly optimistic validation results?

ML Python

from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression

X = [[i] for i in range(100)]
y = [0]*50 + [1]*50

# Split data
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=42)

# Retrain model
model = LogisticRegression()
model.fit(X_train + X_val, y_train + y_val)

# Evaluate
score = model.score(X_val, y_val)
print(f"Validation accuracy: {score}")

A The random_state is fixed, causing biased splits.

B The test_size parameter is too large, reducing training data.

C The model is not fitted before scoring.

D The model is trained on validation data, causing data leakage.

💡 Hint

Check if validation data is used during training.

The code trains the model on both training and validation data, so validation accuracy is not a true measure of performance.