Pt1420 Unit 1 Problem Solving Paper
5. Let us now assume that the values of n, m, and p stay the same (n = 100, m = 1000, and p = 100). However, the time needed to execute compute_something_expensive_and_non_parallelizable(i, j) varies at random between 1 second and 20 hours. How would you now map the tasks to processors to make the computations efficient?
Solution:
The above problem refers to problem of load balancing where time of execution of each task varies at random. Dynamic mapping technique can be used for solving the above problem. In dynamic mapping technique, tasks are managed by Master node and all other nodes that depend on Master for work are called slave nodes.
Here in the given problem, each iteration of inner loop is taking random execution time. Tasks can be created based on iterations of inner …show more content…
Is the following statement valid? This setup solves the problem of concept drift. Justify your answer. Solution:
Given: An updatable Naïve Bayes model that is re-trained on incoming data, using all the data from starting of the stream.
Solution:
This may solve the problem of concept drift but not effectively.
If the stream is not old and data that is accumulated is not old , then the given updatable Naïve Bayes can have fast adaptation to concept changes and solve problem of Concept drift.
If the stream has started long back, then it accumulates old data along with new incoming data, where the old data tends to become outdated. When model is re-trained on this complete stream’s data, then it reacts to concept changes slower. Training the learner/model with stream data that has old accumulated data can induce inaccuracy in predictions.
Ideal solution is to maintain Sliding Window that stores most recent examples in First In First Out fashion. Small window sizes can assure fast adaptation in times of concept drift but fails in stable period. A large window size reacts slower to concept drift but gives a better performance at stable
Solution:
The above problem refers to problem of load balancing where time of execution of each task varies at random. Dynamic mapping technique can be used for solving the above problem. In dynamic mapping technique, tasks are managed by Master node and all other nodes that depend on Master for work are called slave nodes.
Here in the given problem, each iteration of inner loop is taking random execution time. Tasks can be created based on iterations of inner …show more content…
Is the following statement valid? This setup solves the problem of concept drift. Justify your answer. Solution:
Given: An updatable Naïve Bayes model that is re-trained on incoming data, using all the data from starting of the stream.
Solution:
This may solve the problem of concept drift but not effectively.
If the stream is not old and data that is accumulated is not old , then the given updatable Naïve Bayes can have fast adaptation to concept changes and solve problem of Concept drift.
If the stream has started long back, then it accumulates old data along with new incoming data, where the old data tends to become outdated. When model is re-trained on this complete stream’s data, then it reacts to concept changes slower. Training the learner/model with stream data that has old accumulated data can induce inaccuracy in predictions.
Ideal solution is to maintain Sliding Window that stores most recent examples in First In First Out fashion. Small window sizes can assure fast adaptation in times of concept drift but fails in stable period. A large window size reacts slower to concept drift but gives a better performance at stable