Which of the following is the most significant factor that increases the cost of training a machine learning model in a cloud environment?
Think about what requires more computing power and time during training.
Larger datasets require more computation and memory, which increases cloud resource usage and cost. Simple models and fewer epochs reduce cost, and pre-trained models save training time.
You want to deploy a model for real-time predictions with minimal cost. Which model choice is best to reduce inference cost while maintaining reasonable accuracy?
Consider model size and speed during prediction.
Small decision trees are fast and cheap to run, making them cost-effective for real-time inference. Large or ensemble models increase compute cost.
Which hyperparameter adjustment is most likely to reduce training cost without severely impacting model performance?
Think about how batch size affects training speed and resource use.
Larger batch sizes allow more data to be processed at once, speeding up training and reducing cost. More epochs, low learning rates, or deeper models increase cost.
You have two models: Model A costs $100 to train and achieves 90% accuracy. Model B costs $150 to train and achieves 92% accuracy. Which metric best helps decide if the extra cost is justified?
Consider how to measure value gained per cost spent.
Cost per accuracy improvement shows how much extra money is spent for each percent gain, helping balance cost and benefit.
Given the code below for training a model, which line causes unnecessary cost increase by repeating data loading every epoch?
for epoch in range(10):
data = load_data_from_disk()
model.train(data)
Think about when data loading should happen to avoid repeated work.
Loading data inside the loop causes repeated disk reads, increasing time and cost. Loading once before the loop is more efficient.