thosam1 / bachelorproject Goto Github PK

Whenever you have a plot, write a brief description about what is the main insight from the plot. If there is no insight, there is no need to plot it!

The times at which you eliminate arms, and the probability estimates seem correct. But the plots below show linear regret, which seems wrong

why is the plot for number of times suboptimal arm being played increasing linearly? is a suboptimal arm being played every time?

regret plot is also piecewise linear. I would have expected it to be a bit more smooth.

Draw vertical dashed lines to show the exploration stage of each arm.

Correct the title of the regret plot; it says epsilon greedy, but it should be explore-then-commit. The same issue is there in strategy 2 as well.

The histogram plots will give more information if each time you divide by total number of plays., so that each value is between 0 and 1. Maybe you can also plot the histogram of the true probabilities just beside, so as to get an estimate.

Can you give more details on explore-then-exploit by elimination? What are the parameters you give as input to the algorithm?
What happens in each round? Perhaps you could reuse your functions from strategy 1?

Are you eliminating the arms in the same order as their probability of giving a reward? This is what I meant by writing some explanation below your plots in a previous issue.

In the plot for number of wins per arm as a function of nth play, just cut of the plots for the arms you stop playing, instead of making them continue out flat. perhaps you can replace the remaining numbers by None, and the plot will automatically correct for it.

your current ipynb is rather big, it's a bit difficult to navigate.
you can create different ipy notebooks for different weeks tasks.
common functions can be put in .py files that are imported into the notebook.
The main code and the plots can be kept in the notebook.