第 15 / 88 页
非代码面试题
显示 20 / 1751 道匹配题目
答题状态：未尝试未正确已正确
ID题目领域难度题型进度权限
2410Why Regularization Can Raise Train Error but Lower Test ErrorWhy is it perfectly consistent for regularization to worsen train fit but improve out-of-sample MSE?机器学习中等essay未尝试面试订阅2412Why Model Rankings Can Flip as n GrowsWhy can a simple model beat a flexible one at small n and then lose badly once n is large?机器学习中等essay未尝试面试订阅2413Why Bagging Mainly Targets VarianceWhy is bagging usually described as a variance-reduction tool rather than a bias-reduction tool?机器学习中等essay未尝试面试订阅2414Why Irreducible Noise Caps the Best Achievable Test ErrorWhy can model improvements stall even after both bias and variance seem small?机器学习困难essay未尝试面试订阅2415Why a Stable but Biased Model May Be Preferred OperationallyWhy might a desk prefer a slightly biased model that behaves predictably over a lower-bias model whose outputs swing wildly retrain to retrain?机器学习困难essay未尝试面试订阅2418Why Small Validation Sets Overreact to Complex ModelsWhy can small validation sets make model-comparison results look much noisier for complex models?机器学习中等essay未尝试面试订阅2420Why Deployment Preferences Can Differ From Benchmark-MSE PreferencesWhy can the model that minimizes benchmark MSE fail to be the one a production team actually deploys?机器学习困难essay未尝试面试订阅2421Bayes Act Under Weighted Squared Loss 1Suppose the loss is L(a,Y)=W(Y-a) 2 where W>0 is observed at prediction time. What predictor minimizes E[L(a,Y)|X,W]?机器学习简单derivation未尝试免费2423Weighted Log-Loss Bayes Probability 3For binary Y with P(Y=1|X)=p, consider weighted log-loss L(q,Y) = -alpha Y ln q - beta (1-Y) ln(1-q). What probability q minimizes the conditional expected loss?机器学习中等derivation未尝试免费2424Convexity of the Log-Cosh Loss 4Show that ell(r)=ln cosh(r) is convex in the residual r.机器学习中等derivation未尝试免费2425Why Asymmetric Loss Changes the Optimal Prediction 21Why does changing the relative penalty on overprediction versus underprediction generally move the Bayes act away from the conditional mean?机器学习困难essay未尝试面试订阅2427Decision Threshold Under Asymmetric Classification CostA false negative costs 5 and a false positive costs 1. If p is the predicted probability of the positive class, above what threshold should you classify as positive?机器学习中等derivation未尝试面试订阅2428Weighted-Absolute-Loss Constant ForecastDemand Y takes values 0, 2, and 5 with probabilities 0.5, 0.3, and 0.2. Under a loss of 3(Y-a) + + 1(a-Y) +, what constant forecast a minimizes expected loss?机器学习中等derivation未尝试面试订阅2429Total Huber Loss on a Residual SetUsing Huber loss with delta = 1, compute the total loss on residuals 0.5, -1.2, and 3.0.机器学习中等数值题未尝试面试订阅2430Why the Weighted-Brier Bayes Act Is Still a Weighted Mean 7For binary Y and weighted squared loss alpha Y (1-q) 2 + beta (1-Y) q 2, derive the Bayes probability q as a function of p=P(Y=1|X).机器学习困难derivation未尝试面试订阅2433Pinball Loss Subgradient at the Kink 9For pinball loss rho tau(r)=tau r if r>=0 and (tau-1)r if r<0, what is the subgradient set at r=0?机器学习中等derivation未尝试面试订阅2434Why a Strictly Proper Loss Must Prefer Truthful Probabilities 10Why is it desirable for a probabilistic classification loss to be strictly proper rather than merely classification-accurate?机器学习困难derivation未尝试面试订阅2435Why Log-Loss Punishes Overconfidence More Than Brier LossWhy does log-loss react much more harshly than Brier loss when a model assigns near-certainty to the wrong class?机器学习困难essay未尝试面试订阅2437Why Huber Sits Between Squared and Absolute LossWhy is Huber loss often described as sitting 'between' squared loss and absolute loss?机器学习中等essay未尝试面试订阅2438Why Convexity Makes Averaging Predictions SafeWhy does convexity of a loss function support the intuition that averaging similar predictors often cannot hurt too much?机器学习困难essay未尝试面试订阅