Seeds

Seed Control

• Allows seed control across multiple experiments
• Sets seeds for pythons random module, numpy's random num generator, and pytorch
• You can set and get the current state of all random number generators

Seed Selection

• Find a seed (sequence of tokens) from the dataset based on your desired pitch preference, complexity, and length.
• Function that takes a sequence and returns its average pitch, number of unique notes, pitch range, and length of sequence. This is used when searching for a seed that matches the given criteria.
• Another function to compute the values that should be used for ranges. These ranges allow the user to easily select the type of seed they want. It takes a sample of the dataset and essentially figures out what counts as 'low' vs 'medium' vs 'high' pitch, 'simple' vs 'medium' vs 'complex' for complexity, and 'short', 'medium', and 'long' for length. This results in actually being able to find seeds that match what the user has specified.