fig2
![Phthalonitrile melting point prediction enabled by multi-fidelity learning](https://image.oaes.cc/e26ecc19-00f3-4253-8f7b-c1296b9e7eb5/jmi4027.fig.2.jpg)
Figure 2. (A) Gene combination strategy for creating virtual PN molecules. Specific spacers and substituents are combined into the template to generate a vast candidate space; (B) Two-dimensional distribution of the sampling space and candidate space generated by t-SNE analysis. Sampling structures (in yellow) evenly cover the entire candidate space (in green), indicating that the sampling is representative and can avoid selecting too many similar structures. PN: Phthalonitrile; t-SNE: t-distributed stochastic neighbor embedding.