﻿Automatic Analysis Parameters

MODEL CONFIGURATION:
• models_path : Path to the models folder containing trained neural networks
• analyze_with : Model file to use for analysis (.keras format)
• model_bodyparts : Bodyparts used to train the model (subset of all bodyparts)

COLABELS CONFIGURATION:
• colabels_path : Path to the colabels file containing manual annotations
• labelers : List of labelers on the colabels file (as found in the columns)
• target : Name of the target column on the colabels file

DATA SPLIT SETTINGS:
• focus_distance : Window of frames to consider around an exploration event
• validation : Percentage of the data to use for validation (e.g., 0.15 = 15%)
• test : Percentage of the data to use for testing (e.g., 0.15 = 15%)

ANN CONFIGURATION:
• recenter : Whether to recenter the data for neural network input
• reshape : Whether to reshape the data (set to True for RNN)
• reorient : Whether to reorient by rotating coordinates so the south-north vector points upward

RNN_width - Defines the temporal window of the RNN:
  • past : Number of past frames to include in RNN window
  • future : Number of future frames to include in RNN window  
  • broad : Broaden window by skipping frames further from the present

• units : Number of neurons on each layer (list of integers, e.g., [32, 16, 8])

TRAINING PARAMETERS:
• batch_size : Number of training samples processed before updating weights
• dropout : Randomly turn off a fraction of neurons (0.0-1.0, e.g., 0.2 = 20%)
• total_epochs : Each epoch is a complete pass through the training dataset
• warmup_epochs : Epochs with increasing learning rate at the start
• initial_lr : Initial learning rate (e.g., 1e-05 = 0.00001)
• peak_lr : Peak learning rate reached after warmup (e.g., 0.0001)
• patience : Number of epochs to wait before early stopping if no improvement