Adaptive memory records search history so move frequency can influence later decisions. Frequency-based memory tracks how often specific moves occur and can penalize frequently used, unsuccessful, or less promising moves. section 6.2.1
Within tabu search, adaptive memory supports responsive exploration through two complementary records:
- Recency-based memory records recently visited moves and helps prevent immediate revisits. section 6.2.1
- Frequency-based memory records historical move counts across the search and guides longer-term diversification. section 6.2.1 section 6.7
- Together, these records help local search move beyond local optimality while balancing exploration against exploitation. section 6.2 section 6.7
A frequency table for a small route search
In a four-city traveling-salesman search, a tabu structure can store recency values for temporarily forbidden moves alongside frequency counts showing how often each move has been used. The counts provide historical context when selecting later neighbors.
Implementation switch
A generic optalgotools TabuSearch solver exposes frequency-based memory through use_longterm=False by default. Enable the option while tuning max_iter, tabu_tenure, neighbor_size, and aspiration settings. section 6.4
