saxpy库find_best_discord_brute_force函数参数及输入要求咨询

Nice question! Let's break down each parameter clearly so you can use this function effectively for finding discords in your time series:

• series: You don’t need to input SAX words here—this should be your raw time series numerical data (like a Python list or NumPy array of your time-stamped values). The saxpy library handles all the underlying SAX conversion and similarity calculations internally, so just feed in your original dataset directly.

• win_size: This is the length of the sliding window you want to use to extract sub-sequences from your time series. For example, if you’re working with hourly temperature data and want to find anomalous 24-hour periods, set win_size=24. All candidate discords will be continuous sub-sequences of this exact length.

• global_registry: This is a tracking object from saxpy’s registry module (specifically, a GlobalRegistry instance) that stores distance metrics and candidate discord data during the brute-force search. You’ll need to initialize it first (e.g., registry = GlobalRegistry()) and pass it in. Its job is to keep track of the minimum distance between each sub-sequence and all others—since discords are the sub-sequences with the largest minimum distance, this registry helps identify that top anomaly.

• z_threshold (optional, default=0.01): This is a filtering threshold based on Z-score, used to skip unnecessary comparisons with sub-sequences that are too similar to each other. It helps speed up the brute-force search by ignoring trivial, non-anomalous matches.

Quick Example Usage #

from saxpy.registry import GlobalRegistry
from saxpy.discord import find_best_discord_brute_force

# Your raw time series data
raw_time_series = [2.1, 2.3, 2.2, 15.7, 2.4, 2.2, ...]  # truncated for brevity
window_length = 5
registry = GlobalRegistry()

# Find the top discord
best_discord = find_best_discord_brute_force(raw_time_series, window_length, registry)

内容的提问来源于stack exchange，提问作者lord AJ