"I want to make a ** demand forecast ** for each convenience store x product." → Machine learning / statistical modeling-like story Prediction and elucidation of phenomena
"I want to maximize profits ** based on demand forecasts for each convenience store x product." → Mathematical optimization-like story Maximize profits (maximize / minimize KPIs after the phenomenon is elucidated)
Between machine learning and mathematical optimization: There is also the BanBit problem (reinforcement learning)
There are various levels of data science / AI / analytics
Decision-making steps ① Aggregation (what will happen) ② Explanation (why it happened) ③ Prediction (what will happen) ④ Decision making (what to do) ⑤ Action (actual action)
Achievement stage of analytics ① Descriptive analytics ② Diagnostic analytics ③ Predictive analytics ④ Prescription analytics (decision support by AI) ⑤ Prescription analytics (decision-making by AI)
Typically ① Aggregation (what happens) DWH / BI tool ① Descriptive analytics
Opportunity learning / statistical analysis performs (2) explanation (why it happened) and (3) prediction (what happens) ① Descriptive analytics ② Diagnostic analytics ③ Predictive analytics
④ Make decisions (what to do) and ⑤ Actions (actual actions) by mathematical optimization ① Descriptive analytics ② Diagnostic analytics ③ Predictive analytics ④ Prescription analytics (decision support by AI) ⑤ Prescription analytics (decision-making by AI)
| task | Input to the system | System mechanism | System output | Typical technology |
|---|---|---|---|---|
| Forecast / estimation | ⭕️ | Unknown → estimated | ⭕️ | Machine learning / statistics |
| optimisation | The best one is unknown → Search | ⭕️ | I want to minimize / maximize | 数理optimisation |
| Description | Forecast | Decision support | Decision making | |
|---|---|---|---|---|
| Stock Trading (DayTrading) | ○ | ○ | ○ | ○（HFT） |
| Supermarket purchasing plan | ○ | ○ (Demand forecast) | ○ (optimal purchasing) | ○ |
| Rental of construction equipment | ○ | ○ (Demand forecast) | ○ (Optimization of equipment deployment between branches) | ○/?(Conflict of incentives between branches) |
| Commodity trade | ○ | ○ (Price forecast) | ○ (Suggestion of trading timing) | ?? (Evaluation of political risk) |
| Agriculture | ○ | ○ (Harvest forecast) | ○(Fertilizer optimization)/?(sales plan) | ？ |
| M&A | ○ | ?? (High individuality) | ? | ? |
Prone to issue / operation dependence
・ Advertising allocation problem → Maximizing advertising effectiveness (CVR / CTR) ・ Vehicle delivery planning → (weight of one package, optimal route, etc.) ・ Factory production plan → Adjustment of raw materials and production volume ・ Cooperation with other factories, etc. ・ Administrative facility placement problem → (school, etc.) ・ War → (Efficient attack) ・ Portfolio → (Balance of risk / return ・ Risk can be reduced even with the same return) ・ Shift creation → (Adjustment of roles and burdens)
DWH and BI tools https://it-trend.jp/bi/article/bi_dwh
python3 numpy pulp
How to study About 50 optimization problems for application (book)
Recommended Posts