Update regression_forecast method and bug fix

[muazhari]

What I did:

• Update regression_forecast method to make it work with extra features and return MSE.
• Update README.md to add the regression_forecast method usage explanation.
• Update regression_forecast_optimize method sorter bug because of the wrong comparator.
• Update regression_forecast_optimize method slow performance because it has log print.
• Update regression_analysis method bug because it has a circular function call.

What I found:

• I think there is something wrong with the regression_forecast_mse method in this code block below,

  for (j = 0; j < coeffs.length; j++) {
    if (options.method == "ARMaxEntropy") {
      buffer[i][1] -= backup[i - 1 - j][1] * coeffs[j];
    } else {
      buffer[i][1] += backup[i - 1 - j][1] * coeffs[j];
    }
  }

  why it calculated with observed real data (backup)? not the trained ones (buffer)? I think it should be revised like below, is this right?

  for (j = 0; j < coeffs.length; j++) {
    if (options.method == "ARMaxEntropy") {
      buffer[i][1] -= buffer[i - 1 - j][1] * coeffs[j];
    } else {
      buffer[i][1] += buffer[i - 1 - j][1] * coeffs[j];
    }
  }

Note: I don't know and don't have credible knowledge to touch/revise/update/add the core forecasting features. It seems to have some flaw in it that I not aware of and this problem is crucial to the reliability of the library. Please someone if you are able to fix it or contribute something about it, I truly appreciate you.

[26medias]

Seems right

[muazhari]

can you update the readme at the npm website? it still not automatically updated to there