Adding profiling

[JeanPierreMR]

Description of the problem

There is no profiling, and i think its something really important for a data structure, with this we could compare how well they are performing and help us detect when there is something that is slowing down the structure too much.
It doesn't have to be too complicated, a simple test code run by cProfile will do

Example of the problem

References/Other comments

[czgdp1807]

How will that be done? How will we figure out that whether the theoretical complexity of a data structure matches with the profiling data?

[JeanPierreMR]

How will that be done? How will we figure out that whether the theoretical complexity of a data structure matches with the profiling data?

To be able to compare it we could design a test with different sets of random data (varying in the amount of data) and then plot the time it takes for each set in a graph, and that should give us an idea of the practical complexity of the data structure. Depending on the data structure, this could be hard or easy.
Maybe we could use other tools than cProfile, depending on the need of each test

[czgdp1807]

then plot the time it takes for each set in a graph

How can we compare graphs automatically?

[czgdp1807]

We can create performance tests for each data structure. We can test all the features of a data structure by creating sufficiently large test cases for each one of them. The tests should fail if they do not complete within a certain time limit. For example, for O(N) time complexity, 100000 size should be sufficient and the performance tests should complete under 10s(due to python). We can try this thing out on linear data structures as a first step.

[HarsheetKakar]

Instead of them failing we should just let them run, it will give us numbers which will help developers decide which ds to use according to their usecase. It would be nice if we could have kind of a framework which will compare selected algorithms at given number of inputs.

[czgdp1807]

This issue is more relevant now since we are starting to implement C++ backend for our APIs. See, #432

Instead of them failing we should just let them run, it will give us numbers which will help developers decide which ds to use according to their usecase.

I agree.