[ML] Downgrade log severity for a batch of recoverable errors

[darius-vil]

Fixes #2888

[prodsecmachine]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scanner	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.

[darius-vil]

Closes #2888

[edsavage]

LGTM, just the one comment

[valeriy42]

Great start @darius-vil ! 🚀 I left a few improvement suggestions.

Can you also please update the reporting of bad category error in lib/maths/common/CXMeansOnline1d.cc:468 as described in the issue #2888?

You can use a numerical error check along the lines of:

namespace {
bool isNumericalError(const std::exception& e) {
    // Check exception type
    if (dynamic_cast<const std::domain_error*>(&e) ||
        dynamic_cast<const std::overflow_error*>(&e) ||
        dynamic_cast<const std::underflow_error*>(&e)) {
        return true;
    }
    
    // Check message content as fallback
    std::string msg = e.what();
    return msg.find("boost::math") != std::string::npos ||
           msg.find("Scale parameter") != std::string::npos ||
           msg.find("must be > 0") != std::string::npos ||
           msg.find("-nan") != std::string::npos ||
           msg.find("-inf") != std::string::npos;
}
} 

[valeriy42]

LGTM. Great work! 👍