วีดีโอ

Will have to take a look at this, I am imagining the captions are auto-generated… just a shame they have faltered at that specific instance! Glad you liked the video regardless!

@@HysterResearch haha regardless it was a good video

That’s a great question. In this scenario, you are right - it is not necessary and you could have them simply as 3x1 where each position is defined solely by the end point. However, for further problems in the future, we might have a system which has more than one reference frame where the origin of the unit vector coordinates are not zero values. So it’s just good house keeping to keep for the future

Thanks for noticing! You’re absolutely right - it’s a mistake. On line 88 I check a random value against the loudness of the bat, this should be the pulse rate

I am not familiar with this optimisation algorithm, but I will check it out!

For many applications, you won’t notice much difference. Conventional programming and the increase computationally speed we experienced over recent years has began leading to readable code over unnecessarily efficient code. Nevertheless, using Numba for this application does pose a speed advantage, but you have to jump through compatibility hurdles. Essentially, Numba generates C code for execution, meaning that the first run will require generation of the code and subsequent handle calls will have much faster execution. However, if you continue to re-run the compilation, you will not experience any benefit. I am unsure fully of your context, but just ensure you are not re-compiling the code during each call.

@@HysterResearch i was messing around with the possibilities numba offers and was checking how this particular feature worked from that video. What i wanted to use it for, is my implementation of neural network layers, based on different neuron types. The types i wanted to use were taken from the book of one of academic lecturers here in Poland. Those were: simple perceptron based layer (with bias, represented as simple list of variables in pure python), sigmoid based, and radial function neurons (same representation). However when i tried just simple matmul to calculate the impact of the backpropagation, as well as feedforward operations (my end goal and possible speed increase as i thought so) effects were underwhelming to say the least

However one of the challenges i posed to myself was to use numba without relying on the numpy, so in the mode where it would generate pure C code under the hood that is optimized differently and it failed miserably... So my alternative to using numpy still required the knowledge of numpy and the numpy itself which defied its purpose...

There's a nested loop, 1 inner loop with j and 1 outer loop with i which makes the program go through n element of the array n-1 time and fully sort it

Interesting. Are you able to send the code and the error you received?

@@HysterResearch hello, thank you for replying. Would you please give your email, so i can attach other functions willed to be plotted, to be merged within your optimisation_function?

@@HysterResearch my destined function is a relation between two variables, of the form Y = X (not f(x, y)) where only plot(x, y) is needed as the output graph where onto the population will be distributed (unlike surf or contour))

@@HysterResearch if there is any obstacle with providing any email address, i can copy/paste the plot function, so you can modify Optimisation_Function accordingly

Feel free to contact me on codingforscience@gmail.com

Thank you for your questions. Question 1: In the paper the author specifies that the velocity is updated based on the previous velocity, to which the current populations position has the historic best population positions subtracted from it. I found that negating the previous velocity values provided better algorithm robustness and repeatability whilst also aided in convergence. However, by negating the previous velocity, this velocity value now becomes negative, and the bats diverge as opposed to converge. This was the reasoning for switching the position difference around. Questions 2: You are correct in your observation, I have omitted the update. In the paper the if statement has two conditions, if the fitness improves and if a random sample is lower than the loudness value of the bat. To me, this process seemed illogical. Not only does it add an additional stochastic element to the algorithm, it also neglects potential improvements to the global minima based on a randomly generated float relative to the loudness. For these reasons, I opted to neglect it. However, feel free to reverse these changes, and let me know if you have similar success when obtaining a solution.

For demonstration purposes, this problem was conducted on a two-dimensional set of optimisation test functions as this would enable a three-dimensional visual output of the solution. If you have a higher dimension problem, you can increase the column vector of the population to correspond to each additional dimension. You will need to create a bespoke fitness function based around your specific application, and visualising the output is no longer easy or viable. The code itself should be easily adaptable, with the only real change being to the development of your own fitness function, replacing the optimisation test functions used during this demonstration.

@@HysterResearch I also hope u can help to educate me by making this code in multidimensional function. I have try but unsuccessful.

I have asked myself this question multiple times as well. During the diagnostic period of developing this simulation, I found that using prior knowledge of the previous velocity values actually hindered accurate exploration. But you are correct in identifying that for this demonstration the previous velocity has been negated.

th-cam.com/video/WluUkvVXbwg/w-d-xo.html 💐💐💐