FAST experimental data for paper "Applying hybrid clustering in pulsar candidate sifting with multi-modality for FAST survey"

The FAST data set was used to further verify the robustness and efficiency of our proposed algorithm. At first, to change the class imbalance ratio between pulsar and non-pulsar samples in a single block, 1600 known pulsar samples from multiple types were prepared as a known sample set s′ and added to this data set. Some of the pulsar samples in s′ are known pulsars searched by CRAFTS during synchronization testing, and they can be found in the linked star catalog:http://groups.bao.ac.cn/ism/CRAFTS/CRAFTS/. More over, the dispersion measurement (DM) range of these pulsar samples is often set to [2,1000](cm−3pc). The remaining in s′ were collected from foreign surveys such as HTRU. All the new candidates in FAST data (containing the above known sample set s′ ) were formed by the DCA fusion of the 1-D feature arrays and related 2-D TVP arrays, which were extracted from corresponding pfd files by the feature extraction programs of HTRU2 and PICS. Moreover, the known sample set s ′ (1600 real pulsar samples) and 10000 non-pulsar samples were used for DCA algorithm training. Next, the sliding window size was also set to Batchsize = 2 and the unit-size was 1251, then the original data set was divided into (g1, g2, · · ·, g125, g126) , where g1, g2, · · ·, g125 = 1251 but g126 = 1241 . As a result, the experimental data set consisted of 126 data blocks, i.e.Block(1):[ s′ , g1, g2], Block(2):[ s′ , g2, g3],..., Block(125):[ s ′ ,g125, g126], Block(126):[ s′ , g126, g1]. Note that, the original 78 pulsar samples from FAST data were randomly distributed in (g1, g2, ···, g125, g126).