了解複雜系統中因果關係的重要一環是如何檢測資訊流的方向。轉移熵（transfer entropy）被視為一種檢測資訊流的方法。然而，傳統方法中使用轉移熵需要事先確定歷史 長度，而不同的歷史長度可能導致截然相反的結果。尤其對於預測系統而言，為了獲得正確的資訊流方向，轉移熵需要更長的歷史長度。然而，過長的歷史長度可能導致轉移熵出 現巨大偏差，使結果難以解讀。在本研究中，我們提出了一種基於壓縮演算法估計轉移熵的新方法，稱之為壓縮轉移熵。這種方法可以在不需要事先指定歷史長度的情況下檢測資訊流方向。我們建立了兩個具有單一資訊流方向的模型，並使用真實的斑馬魚數據進行測試，以評估壓縮轉移熵的檢測能力，同時與傳統轉移熵方法進行比較。研究結果表明，壓縮轉移熵相較於傳統方法，在不需要指定歷史長度的情況下能夠準確地獲得結果。;Understanding the causality relationships in complex systems is an important aspect of determining the direction of information flow. Transfer entropy is regarded as a method for detecting information flow. However, traditional approaches using transfer entropy require determining the history length beforehand, and different history lengths can lead to contradictory results. This is particularly challenging for anticipation systems, as transfer entropy requires longer history lengths to obtain the correct direction of information flow. However, excessively long history lengths can introduce significant bias in transfer entropy, making the results difficult to interpret. In this study, we propose a new method called ”compressed transfer entropy” that estimates transfer entropy based on compression algorithms. This method enables the detection of information flow direction without the need for specifying a history length in advance. Two models with unidirectional information flow, along with real-world zebrafish data, were utilized to evaluate the detection capability of compressed transfer entropy, comparing it with traditional transfer entropy methods. The results demonstrate that compressed transfer entropy provides accurate results without the requirement of specifying a historical length.
