Settings

Global settings

You can set the global settings for run in the file data/global_settings.json. The available settings are:

• "sample_dir": "/foo/bar"

With this value one sets the location where the sample is located. Needed only for the MNIST numbers dataset currently

---

• "nthread": 28

With this one sets how many threads are to be used during the computation

---

• "output_dir": "$HOME/foobar"

Directory where the results will be saved. Environment variables like $HOME can be used

---

• "sample_type": "tth"

The sample type that is to be calculated. Options are either "tth" or "mnist"

---

• "num_classes": 2

Number of different classes where the event can be classified. For "tth" it needs to be 2 (signal or background) and for "mnist" it is 10 (10 numbers)

---

• "bdtType": "evtLevelTT_TTH"

---

• "trainvar": "noHTT"

---

• "channel": "3l_1tau"

---

• "fitness_fn": "d_roc_1.5"

The fitness function used in the optimization. Available options are: 'f1_score_test' , 'g_score_test' , 'test_auc'

• "trainvar_opt": 0

0 if no training variable optimization is wanted. Choose 1 to perform training variable optimization. Not available for "mnist" samples.

• "ml_method": xgb Available options: 'xgb' and 'nn'

Particle swarm optimization settings

The settings for particle swarm optimization can be found under data/pso_settings.json. The available settings are:

• "iterations": 2

Maximum number of iterations that the particle swarm does.

---

• "sample_size": 3

Number of "particles" that are trying to find the optimum.

---

• "compactness_threshold": 0.1

How compact the parameters of all particles need to be together. This is calculated using the mean of coefficient of variation.

---

• "w_init": 0.9

Initial inertial weight. This is set higher than the final inertial weight in order to make the particles converge in the global optimum faster.

---

• "w_fin": 0.4

Final inertial weight that is reached with the final (maximal) iteration.

---

• "c1": 2

Cognitive weight of each particle. The bigger it is, the more it tends to go in the direction of the personal best parameters.

---

• "c2": 2

Social weight of each particle. The bigger, the more it tends to go in the direction of the global best parameters.

Genetic algorithm settings

The settings for genetic algorithm can be found under data/ga_settings.json. The available settings are:

• "sample_size": 3

Number of "individuals" being mutated.

---

• "sub_pops": 1

Number of sub-populations that the individuals are divided into.

---

• "iterations": 2

Maximum number of iterations the algorithm will perform.

---

• "threshold": 0.001

Threshold of improvement of the current generation compared to the previous one. If the improvement percentage is equal to or lower than the threshold for at least two consecutive generations, the algorithm will stop iterating.

---

• "mut_chance": 0.03

The probability of a mutation occurring in a parameter value.

---

• "elites": 1

The number or percentage of best (elite) individuals to preserve from the previous generations.

---

• "culling": 0

The number or percentage or worst individuals to remove (cull) from the current generation and replace with random new ones.