-
What is the general trend in the curve?
The trend is sort of asymptotic, seeming to converge towards 90%. When the train/test split is 50%, the best result is about 80% test accuracy, which I think is pretty good. -
Are there parts of the curve that appear to be noisier than others? Why?
The noise seems pretty evenly distributed even across different test/train percentage splits by my findings. -
How many trials do you need to get a smooth curve?
Usually around 100 runs are needed, although 1000 is even nicer. With low C-values, it isn't possible to get rid of noise. -
Try different values for C (by changing LogisticRegression(C=10** -10)). What happens? If you want to know why this happens, see this Wikipedia page as well as the documentation for LogisticRegression in scikit-learn.
I graphed the results with several different C values (10**-10, 10**-9, etc.) and found that the higher the C value was, the stronger the convergence and the higher the test accuracy.
Results:
| 100 runs | 10 runs | 1 run |
|---|---|---|
 |
 |
 |