Unverified50% confidenceFactExact time
Training starts with sparse rewards, and then iterates through reward shaping via an event frequency table; in the event frequency table, movement penalties account for 48.3% of all events, exploring tiles accounts for 21.7%, and attacking monsters accounts for 14.5%.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWith a fixed learning_rate, when batch_size scales linearly with the number of environments and num_epoch remains unchanged, the total displacement pushing the policy toward a new distribution in a single iteration may increase, manifesting as violent oscillation or even collapse of rewards in the early stages of training.68% similarUnverifiedSpecific reward and penalty clauses: analysis +10 for justifiably killing a weak candidate, scouting +8 for reporting a week with nothing to show; unnecessary escalation -15, run crash -10, timeout -5; fabricated data -200 and flagged for review.66% similarUnverifiedA sampling profiler periodically captures call stack snapshots to estimate hot spots, with overhead typically only a few percent, whereas a tracing profiler hooks into every call and return event, with overhead sometimes several times higher.64% similarUnverifiedThe ε-greedy policy selects actions randomly with probability ε and selects the action with the highest current Q-value with probability 1-ε; ε is typically set to 1.0 at the beginning of training and then linearly decayed to 0.01 or 0.1 over millions of frames.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1048204API
curl https://kongchang.com/api/v1/knowledge/claims/1048204MCP
get_claim(id=1048204)