Implemented partial sum cache for SMAIndicator

[sgflt]

Fixes #1139.

Performance of partial sums.

Changes proposed in this pull request:

• longer barSeries benefits from reusage of precomputed partial sums

• random access fallbacks to slowPath

• added an entry with related ticket number(s) to the unreleased section of CHANGES.md

[sgflt]

Test on real data:
Before:
ta4jexamples.backtesting.MovingAverageCrossOverRangeBacktest -- Back-tested 2145 strategies on 6150-bar series in 72520 ms

After:
ta4jexamples.backtesting.MovingAverageCrossOverRangeBacktest -- Back-tested 2145 strategies on 6150-bar series in 6222 ms

[TheCookieLab]

Should provide a lot of value as this is one of the most commonly used indicators. Does this work as expected on a moving bar series?

[sgflt]

I will add edge case tests, to prove functionality.

[sgflt]

Umm, what is expected behavior for moving series?

If I add following test:

    @Test
    public void usingBarCount3UsingClosePriceMovingSerie() {
        data.setMaximumBarCount(13);
        data.addBar(new MockBar(5., numFunction));

        Indicator<Num> indicator = getIndicator(new ClosePriceIndicator(data), 3);

        assertNumEquals((0d + 0d + 2d) / 1, indicator.getValue(data.getBeginIndex() + 0));
        assertNumEquals((0d + 2d + 3d) / 2, indicator.getValue(data.getBeginIndex() + 1));
        assertNumEquals((2d + 3d + 4d) / 3, indicator.getValue(data.getBeginIndex() + 2));
        assertNumEquals((3d + 4d + 3d) / 3, indicator.getValue(data.getBeginIndex() + 3));
    }

It fails on second assert (0d + 2d + 3d) / 2 even with previous implementation.

Expected :2.5
Actual   :2.3333333333333335

data.getBeginIndex() + 0 returns bar at index 0 because i - removedBars < 0;
data.getBeginIndex() + 1 returns also bar at index 0 because i - removedBars == 0;

It seems unstable bars are really unstable and should not be tested?

[sgflt]

Discovered RunningTotalIndicator. I may use this indicator as partialSum and optimize RunningTotalIndicator.

[sgflt]

And now I have discovered PreviousValueIndicator. I think I may combine them too.

[sgflt]

Nope with indexes.

[TheCookieLab]

Yes, you'll want to handle unstable indexes as unstable and assert accordingly.

[TheCookieLab]

@sgflt please run the formatter (mvn -B clean license:format formatter:format test) and re-push

[TheCookieLab]

@sgflt I recently cut the 0.16 release and started 0.17. Can you update your CHANGELOG entry accordingly? Thanks.

[sgflt]

Done