Add SrCnn Anomaly Detection algorithm

[mengaims]

New Algorithm

Implement SrCnn anomaly detection algorithm of KDD 2019 paper "Time-Series Anomaly Detection Service at Microsoft". Onboard Spectral Residual(SR) step firstly.

Benchmark report

1. Dataset

We evaluate on the Yahoo timeseries dataset, which has 367 timeseries and 572966 points in total.

2. Evaluation method

We calculate the Precision, Recall, and F1 score using the method of： https://github.com/iopsai/iops/tree/master/evaluation

3. Score and Latency:

We ran the three algorithms on a machine with Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz, 16GB memory, x64 operating system.

Algo	Precision	Recall	F1	#TruePositive	#Positives	#Anomalies	Average latency to predict the whole dataset	Fine tuned parameters
SSA (need training)	0.582	0.585	0.583	2290	3936	3915	5595ms (training time not included)	Confidence=99, PValueHistoryLength=32, Season=11, and use half the data of each series to do the training.
IID	0.668	0.491	0.566	1924	2579	3915	6163ms	Confidence=99, PValueHistoryLength=56
SR	0.601	0.670	0.634	2625	4370	3915	20930ms	WindowSize=64, BackAddWindowSize=5, LookaheadWindowSize=5, AveragingWindowSize=3, JudgementWindowSize=64, Threshold=0.45

4. Some experiences for parameter tuning.

SSA: Sensitive to the season value, need to find the period of timeseries if seasonality exists. Very sensitive to confidence, we keep it above 98.
IID: Very sensitive to confidence, we keep it above 98.
SR: WindowSize and Threshold are the most important parameters, you can focus on adjusting these two parameters to get a good score. And then tuning JudgementWindowSize will also help.