Unverified50% confidenceConceptExact time
In the evaluation context, sampling bias refers to the composition of an evaluation set deviating from the real deployment distribution due to human selection, causing systematic distortion of scores; practices to avoid it include freezing the evaluation set before training, stratified sampling according to the real distribution, and retaining a holdout set that never participates in tuning.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
Verified若用含测试集的全量数据计算均值和标准差，会导致预处理泄漏，使评估指标虚高，无法真实反映模型泛化能力79% similarVerifiedDrift detection involves a trade-off: setting the threshold too sensitively leads to frequent false alarms and unnecessary retraining, while setting it too loosely may allow model performance to continuously degrade without being detected.76% similarUnverifiedIn dense retrieval, in-batch negative training requires batches to contain informative negatives; random sampling causes the training loss to converge quickly but results in poor generalization performance.75% similarUnverified留出集是从数据中划出一部分在整个开发周期内不用于调优只用于最终验收，目的是防止评估集污染导致评分虚高但泛化能力未提升75% similarUnverifiedCluster-robust standard errors are used to handle cases where samples are not independent: when multiple records correspond to the same case, computing variance as if samples were independent underestimates the standard errors and inflates significance, and the correct approach is to aggregate by cluster first and then adjust the variance estimate.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1101024API
curl https://kongchang.com/api/v1/knowledge/claims/1101024MCP
get_claim(id=1101024)