第 18 / 32 页
非代码面试题
显示 20 / 622 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
4208Larger Gamma in an RBF KernelA soft-margin SVM keeps ||w|| 2 and total hinge loss fixed, but C increases from 0.5 to 1.5 while total hinge loss is 2.0. How much does the objective increase?机器学习中等derivation未尝试面试订阅4209Smaller Gamma in an RBF KernelAt a test point, one support vector contributes +0.9. If its dual coefficient is halved and everything else stays fixed, what contribution remains from that support vector?机器学习中等derivation未尝试面试订阅4210When Kernels Beat Manual Feature LiftsA linear SVM score at a point is 1.2 before feature rescaling. If the relevant feature values are all doubled while w is held fixed, what new score contribution comes from that linear term?机器学习中等derivation未尝试面试订阅4211Why Only Support Vectors Matter at Prediction TimeIf your features are extremely sparse one-hot indicators with huge dimension, would you usually try a linear SVM or an RBF SVM first, and why?机器学习中等essay未尝试面试订阅4212Why the Dual Is Attractive in Kernel SVMsWhy is feature scaling especially important before using an RBF SVM?机器学习中等essay未尝试面试订阅4213Why Hard Margin Is FragileWhy do only support vectors matter at prediction time in an SVM?机器学习中等essay未尝试面试订阅4214Why Kernels Are Not FreeWhy can choosing C extremely large increase overfitting risk in a soft-margin SVM?机器学习中等essay未尝试面试订阅4215A Fast Sanity Check for Kernel-SVM AnswersWhy can a high-degree polynomial kernel become numerically and statistically awkward on unscaled features?机器学习中等essay未尝试面试订阅4216Normalized MDI Share 1A random forest reports total mean-decrease-in-impurity contributions spread=0.42, imbalance=0.21, id bucket=0.07. What are the normalized importance shares, and which feature ranks first?机器学习简单数值题未尝试面试订阅4217Normalized MDI Share 2A model has baseline validation AUC 0.62. After permuting three features separately, AUC becomes 0.57 for value signal, 0.60 for momentum, and 0.61 for zip code. What permutation-importance drops do these imply, and which feature ranks first?机器学习简单数值题未尝试面试订阅4218Normalized MDI Share 3A sector feature is represented by three one-hot columns with impurity-gain importances 0.04, 0.03, and 0.01. Two other features have importances 0.05 and 0.07. If you aggregate the one-hot block into a single group, what are the normalized group shares and which group ranks first?机器学习简单数值题未尝试面试订阅4219Normalized MDI Share 4Two trees contribute split gains to features A and B. Tree 1 contributes A=12, B=5. Tree 2 contributes A=8, B=10. What are the total normalized gain importances for A and B?机器学习简单数值题未尝试面试订阅4220Normalized MDI Share 5A model has baseline log loss 0.400. After permuting feature X, log loss rises to 0.455; after permuting feature Y, it rises to 0.420. What are the permutation importances under a log-loss metric, and which feature is more important?机器学习简单数值题未尝试面试订阅4221Grouped Permutation Drop Pattern 1A model starts at validation accuracy 0.82. Permuting feature X1 alone lowers it to 0.79, permuting X2 alone lowers it to 0.8, and permuting them together lowers it to 0.7. What are the three drops, and what pattern does that suggest?机器学习中等数值题未尝试面试订阅4223Grouped Permutation Drop Pattern 3A model has baseline AUC 0.70. With a correlated twin present, permuting feature A drops AUC to 0.64. After removing the twin, permuting A drops AUC to 0.58. By how much did feature A's permutation importance increase?机器学习中等数值题未尝试面试订阅4224Grouped Permutation Drop Pattern 4An impurity-based feature ranking is id hash=0.40, signal 1=0.35, signal 2=0.25. After limiting max depth, id hash gain is cut in half while the other raw gains are unchanged. What are the new normalized shares?机器学习中等数值题未尝试面试订阅4225Grouped Permutation Drop Pattern 5A feature's permutation importance is measured as baseline accuracy minus permuted accuracy. Under validation set A the numbers are 0.80 and 0.78; under noisier validation set B they are 0.74 and 0.72. What is the relative drop as a percentage of baseline in each case?机器学习中等数值题未尝试面试订阅4226High-Cardinality ID TrapA random forest says a hashed customer ID is the most important feature by impurity decrease, even though the validation permutation drop is almost zero. What is the most likely trap?机器学习中等derivation未尝试面试订阅4227Leakage Proxy TrapA model ranks a 'days since settlement' field as highly important in a fraud predictor, but that field is only known after the case outcome becomes visible. What is wrong with reading that as genuine predictive importance?机器学习中等derivation未尝试面试订阅4228Proxy Feature TrapA tree model gives most importance to ZIP code instead of the underlying income and region variables. Why should you be cautious before concluding ZIP code is the true driver?机器学习中等derivation未尝试面试订阅