VarVisitor isn't numerically stable

[adrian17]

Given example code:

    std::size_t   SIZE = 1000;
    auto data = gen_normal_dist<double, ALIGNMENT>(SIZE);

    //for (auto &value : data) {
    //    value += 100000000.0;
    //}

    MyDataFrame df;
    df.load_data(
        MyDataFrame::gen_sequence_index(0, SIZE, 1),
        std::make_pair("value1", data));

    VarVisitor<double, time_t>  ln_vv;
    df.visit<double>("value1", ln_vv);
    std::cout << ln_vv.get_result() << std::endl;

The result is correct - close to 1.0. Once I uncomment the commented lines, I expected the result to stay ~1.0, since the variance didn't change. Instead, the program shows results like

-8.2002
-41.001
10.2503
53.3013

This degree of error isn't present in most other statistical packages I've seen.

(I didn't test other functions, but I'm assuming they could also use an audit for correctness.)

[hosseinmoein]

I will take a look, thanks

[hosseinmoein]

This has been fixed in master branch