Random Kaggle 20.01.2023|Competitions on LK H3 and Recommender Systems.
Join us: https://discord.gg/sZkePhaWSZ. --- The conversation revolved around two competitions. In the first competition, participants discussed various topics such as the use of LK H3 and diagonal movement, as well as the calculation of distance using a formula. They also talked about the need to input a matrix and set constraints for the library used in the competition. The participants also discussed the use of heuristics and genetic algorithms in solving problems, the importance of caching, and the need to prioritize certain features and use a cost function to determine the best set of features to use. In the second competition, the participants discussed a recommender system where the task was to predict clicks, additions to the cart, and orders. They talked about the importance of predicting orders, which accounted for 60% of the task. The conversation also touched on the need to create models based on user behavior and the possibility of using three models for clicks. The participants also discussed the use of seasonality to predict the popularity of certain products in a recommender system competition. They also talked about the need to check that the number of orders does not exceed the number of items in the cart and the possibility of using metrics to measure similarity between users. The conversation also touched on the use of quadrants in the first competition and the need to add constraints and penalties for certain actions, such as an agent going out of bounds. The participants mentioned the use of checkpoints and configuration resets in their approach. They also discussed a genetic algorithm used to solve a Traveling Salesman Problem with additional constraints. Towards the end of the conversation, the participants discussed the strategies used by a Japanese participant who used Concord software and a genetic algorithm to solve the problem in the first competition. They also talked about the concept of a significant user in the second competition and the need to cluster users based on their behavior on the website. The conversation also touched on the possibility of recommending products based on the frequency of their purchase and the need to exclude products that are constantly sold in any cart. In addition, the conversation delved into the topic of a recommender system and the need to recommend products based on similar cart items. The participants discussed the use of event data from the test set to enrich the data and the possibility of clustering similar cart sequences. They also talked about the importance of recommending products based on user behavior and the need to avoid recommending products that were already in the cart. The conversation also touched on the use of statistics to calculate user features and the importance of time series data in the recommender system. The participants also discussed the concept of user features and how they can be used to group sessions based on user behavior.
Join us: https://discord.gg/sZkePhaWSZ. --- The conversation revolved around two competitions. In the first competition, participants discussed various topics such as the use of LK H3 and diagonal movement, as well as the calculation of distance using a formula. They also talked about the need to input a matrix and set constraints for the library used in the competition. The participants also discussed the use of heuristics and genetic algorithms in solving problems, the importance of caching, and the need to prioritize certain features and use a cost function to determine the best set of features to use. In the second competition, the participants discussed a recommender system where the task was to predict clicks, additions to the cart, and orders. They talked about the importance of predicting orders, which accounted for 60% of the task. The conversation also touched on the need to create models based on user behavior and the possibility of using three models for clicks. The participants also discussed the use of seasonality to predict the popularity of certain products in a recommender system competition. They also talked about the need to check that the number of orders does not exceed the number of items in the cart and the possibility of using metrics to measure similarity between users. The conversation also touched on the use of quadrants in the first competition and the need to add constraints and penalties for certain actions, such as an agent going out of bounds. The participants mentioned the use of checkpoints and configuration resets in their approach. They also discussed a genetic algorithm used to solve a Traveling Salesman Problem with additional constraints. Towards the end of the conversation, the participants discussed the strategies used by a Japanese participant who used Concord software and a genetic algorithm to solve the problem in the first competition. They also talked about the concept of a significant user in the second competition and the need to cluster users based on their behavior on the website. The conversation also touched on the possibility of recommending products based on the frequency of their purchase and the need to exclude products that are constantly sold in any cart. In addition, the conversation delved into the topic of a recommender system and the need to recommend products based on similar cart items. The participants discussed the use of event data from the test set to enrich the data and the possibility of clustering similar cart sequences. They also talked about the importance of recommending products based on user behavior and the need to avoid recommending products that were already in the cart. The conversation also touched on the use of statistics to calculate user features and the importance of time series data in the recommender system. The participants also discussed the concept of user features and how they can be used to group sessions based on user behavior.