园林历史研究中的量化及分析算法研究——以南京明、清杏花村地块为例

【目的】基于园林历史研究中大量私家园林历史信息数据化及数据分析问题,探索比传统研究范式在样本覆盖率和信息复杂降低程度两个方面更佳的研究路径。【方法】以明、清时期南京杏花村地块的45处私家园林为研究对象,通过梳理、验证历史信息,将45处园林、57项独立特征转化为合计2 565项的历史信息判定矩阵。对信息矩阵进行K-means聚类和主成分分析,并将聚类和降维之后的结果从样本覆盖率和信息复杂降低程度两个方面,与传统研究范式,如通过四要素、园主人等分类后提取高频特征研究的方法进行比对,同时比较了传统、聚类、主成分分析所得历史信息规律的差异。【结果】分析比较指出,传统分类后高频的研究方法样本覆盖率一般为45%,信息复杂度约为原样本的70%;指定分类数为5的K-means聚类算法样本覆盖率约为42%,信息复杂度约为原样本的63%;而采用主成分分析法可以跨过分类步骤,得到合计70%以上、信息复杂度为原样本44%的规律。在分类后获得的历史性规律中,采用传统方法只能得到离散信息,以K-means聚类后可以得到跨类型的规律,而用主成分分析法不但能得到跨类型规律,还具有提示性强、特征明显的优点。【结论】南京杏花村地块历史园林发展脉络中,园林的要素和风格更多受地块内已有园林形态影响,而非园主人身份或造园风尚引导,这突破了已有范式的常见结论。在园林历史研究中,面对大量私园时,主成分分析法具有样本覆盖率高、信息复杂度低的优点,是一种可参考的研究路径。

Algorithm and quantization in historical garden research in Xinghua Village,Nanjing, in the Ming and Qing dynasties

【Objective】This article focuses on the digitization of historical information and data analysis of private gardens. Two new research paradigms were explored and compared with the traditional processes in terms of sample coverage and information complexity reduction. 【Method】 Forty-seven private gardens in the Xinghua Village plot in Nanjing during the Ming and Qing dynasties were examined. First, 45 gardens and 57 variables were transformed into a total of 2 565 items in a historical information matrix based on the detailed historical scrutiny. Subsequently, the information matrix was subjected to K-means clustering and the principal component analysis (PCA). Two properties, the sample coverage and information complexity reduction, were studied and compared with those of the traditional research, such as four elements, host identity, and a high-frequency feature analysis. Finally, the differences in historical laws after the above analysis were compared. 【Result】 The traditional, high-frequency analysis had a sample coverage of approximately 45% and retained 70% of the original information complexity; for the K-means clustering algorithm with a specified number of 5, the sample coverage was 42%, and 63% of the original information complexity was retained. In contrast, PCA achieved more than 70% sample coverage and only 44% of the original information complexity. Of the historical laws obtained, the traditional method could procure discrete independent information. However, both K-means clustering and PCA yielded few intersectional laws. In particular, the PCA rules showed strong suggestions and characteristics as obvious advantages. 【Conclusion】The PCA results showed that, in the context of the historical garden evolution in the Xinghua Village plot, garden elements and styles were more affected by the existing garden morphology, rather than the identity of the owners or fashions. This supersedes the common results in the existing paradigm. In conclusion, in the research of large number of private historical gardens, PCA offers the advantages of high sample coverage and low information complexity and can be considered as a standard method.