Normalization

[dmitrii-codes]

1) Two parameters were added:

• AzureML to use a web deployed model on Azure ML (API settings are required).
• Normalized to pick only normalized and important features. New vectors (weights) were obtained;

  To normalize every feature, I used the data binning technique. It is a way to group widely distributed (non-linear) features into a smaller number of "bins" (100 by default). https://en.wikipedia.org/wiki/Data_binning To get rid of unimportant features, I only picked those features where number of unique buckets is more than some specific number (I set 3 by default, in the future it is recommended to compare other values). The full report can be provided upon request. 2) settings.json is used to store the feature bins, indexes of the important features and API settings, as well as the weights and parameterized threshold; 3) Minimum version of the Windows PowerShell host was lowered to 1.8.3 so users could use VS Code debugger. 4) Fixed "empty script" bug -> empty script is not obfuscated a priori, no errors need to be thrown.

[edyoung]

These changes were implemented by Dmitrii as part of work to improve the accuracy of Revoke-Obfuscation on modules submitted to the PowerShell Gallery. We'd like to contribute them upstream. @LeeHolmes

[edyoung]

@danielbohannon can you give us an idea whether you'll be open to merging this kind of thing?