Object Oriented Scientific Programming with C++

There was a small error in the clarification of the Neural Network project description. The mapping of 2d to 1d indices must read

data[ncols*irow+icol]

The deadline for the final project has been extended to 15.2.2022, 23:59 (CET). That also means the we will not run the full spec tests and also not have the final grades before that date. Please contact me by email in case you urgently need a grade before that date.

We offer extra Q&A sessions (via Discord) to help you through the project on the following dates:

• Mon 17.1. 13:45-15:30
• Tue 18.1. 10:45-12:30
• Mon 7.2. 13:45-15:30
• Tue 8.2. 10:45-12:30

The project description has been extended at several places to clarify issues that some of you have raised during the lab sessions. These are:

• Clarification of the MSE loss function (What is \(n\)?, What is the format of \(y_\text{pred}\) and \(y_\text{true}\)?, …)

• The get_accuracy(...) function should return values in the range 0-1 and not 0-100 (as percentages).

• The Matrix class requires two more functions

  int getRows() const {...};
  int getCols() const {...};
  

  that return the number of rows and columns, respectively.

There is a small bug in the Neural Network project. The signature of the Matrix constructor must be

Matrix(int rows, int cols, const std::initialize_list<T>& list)

instead of

Matrix(int rows, int cols, const std::initialize_list<T>* list)

The assignment text has been adjusted.

Dear all,

the demo Return type deduction from week 5 contained an implementation for C++17 that used the auto keyword as a replacement for templates, e.g.,

auto sum(auto a, auto b)
{
  return a+b;
}

This is not a C++17 feature (as it was written in the comment) but an extension of the GCC compiler (and only of that). I would not recommend to use this feature because then your code is not portable to another compiler. It also did not make it into the C++17 standard.