-3

Is there a way to programmatically analyze charts and understand if they fit to any of the pre-defined patterns. I was looking to analyze patterns as folows:

enter image description here

I've tried building an algorithm that measures the distances between the various line points, looks ahead and verifies if it matches any of the patterns by keeping a margin of +/- 2%.

I've recently read about scikit-learn and I was thinking if Machine learning can be applied to it, to build on patterns from sample data provided. I'm open to any programming language that can help me achive this. Any inputs would be appreciated.

Akash
  • 4,956
  • 11
  • 42
  • 70

3 Answers3

0

I think this is just the main problem to be solved with Wikipedia - Linear regression

I found this post:

EDIT

hiropon
  • 1,675
  • 2
  • 18
  • 41
  • Would it be possible for you to give me a example on to how to go about applying this for my use case? As I'm not that familiar with it – Akash Oct 16 '17 at 04:37
0

You can use ML-approaches, for example, classifying or clustering this sequences, applying some pre-function (like rate - to unify charts).

But my advice: go to statistics and use time series decomposition: look here - Python approach, the same function you can find in R

It may help you also to divide TS with mix of trends (one-by-one) into different TS.

avchauzov
  • 1,007
  • 1
  • 8
  • 13
0

You can try dynamic time warping (DTW) https://en.wikipedia.org/wiki/Dynamic_time_warping.

Artem
  • 79
  • 9