Chronos: Evaluator enhancement

[TheaperDeng]

Background

Evaluation has become an important topic in our engagement with customers. There are many aspects we need to enhance our evaluator.

• Acurracy
  • Quantitative: Currently we have Evaluator.evaluate to cover this function.
  • Qualitative: We need to provide a plot API for our users to check the forecasting/anomaly detection quality.
• Performance
  • We need an API to evaluate the time cost of a workflow
• Usability
  • We need to support both torch and tf users in evaluator.
  • We need to make the import line simplier.

Methodology

• Acurracy
  • depend on matplotlib (will define the API)
• Performance
  • Evaluator.get_latency (detailed design will be shown in PR.)
• Usability
  • There are 2 possible methods 1. Use numpy ndarray to rewrite most of the metrics 2. Use https://www.tensorflow.org/api_docs/python/tf/keras/metrics to support tf backend (after discussion with @liangs6212 we both believe choice 1 is better).
  • change to from bigdl.chronos.metric import Evaluator.