''

HPMSer - Hyper Parameters Search Tool

---

HPMSer is a tool for searching optimal hyper-parameters of any function. Assuming there is a function:

def some_function(a, b, c, d) -> float

HPMSer will search for values of a, b, c, d that MAXIMIZE the return value of the given function.

To start the search process, you will need to create an object of the HPMSer class by providing to its __init__:

• a func (type)
• parameters space definition passed to func_psdd (with PSDD - check pypaq.pms.base.py for details)
• if some parameters are known constants, you may pass their values to func_const
• configure devices, n_loops and optionally other advanced HPMSer parameters

You can check /examples for sample run code. There is also a project: https://github.com/piteren/hpmser_rastrigin that uses HPMSer.

---

HPMSer implements:

• smart search with evenly spread out quasi-random sampling of space
• parameters space estimation with regression using SVC RBF (Support Vector Regression with Radial Basis Function kernel)
• space sampling based on current space knowledge (estimation)

HPMSer features:

• multiprocessing (runs with subprocesses) with CPU & GPU devices using the 'devices' parameter - check pypaq.mpython.devices for details
• exception handling, keyboard interruption without a crash
• automatic process adjustment
• process saving & resuming
• 3D visualisation of parameters and function values
• TensorBoard logging of process parameters

---

If you have any questions or need any support, please contact me: [email protected]