Table 3 Parameter setting for the experimentations. The T-Trees parameters were tuned according to the experiments and remarks reported in [36]. In the hybrid FLTM / T-Trees approach, k was automatically adjusted to the current cluster size. The values of the FLTM parameters were set after indications from [32], except for card max and δ which were tuned according to our own experience. The cardinality of a latent variable L is computed as an affine function of the number of child SNPs, n c : card(L)=min(α+β×n c , card max ). To avoid questionable empirical setting of DBSCAN’s R parameter, for each of the 14 datasets analyzed, we ran the FLTM learning algorithm for a wide range of possible values of R. For each dataset, we retained the FLTM model with the R parameter that optimized an entropy-based criterion. The value of the other DBSCAN’s parameter, N min , was set to the minimum
Method | Parameter | Description | Value |
|---|---|---|---|
T-Trees and hybrid approach | Â | Size for the blocks of contiguous SNPs (T-Trees) | 20 |
| Â | T | Number of meta-trees in the random forest | 1000 |
| Â | S n | Threshold size (in number of observations), to control meta-tree leaf size | 2000 |
|  | S t | Threshold size (in number of meta-nodes), to forbid expanding a meta-tree beyond this size | ∞ |
| Â | K (T-Trees)K (hybrid) | Number of contiguous blocks of SNPs, or number of clusters in LDMap, to be selected at random at each meta-node, to compute its cut-point | 1000 |
| Â | s n | Threshold size (in number of observations), to control embedded tree leaf size | 1 |
| Â | s t | Threshold size (in number of nodes), to forbid expanding an embedded tree beyond this size | 5 |
| Â | k | Number of variables in a block (T-Trees) or cluster (hybrid), to be selected at random, at each node, to compute its cut-point | size of block or of cluster |
FLTM | α | Three parameters to model the cardinality of each | 0.2 |
|  | β | latent variable as an affine function with a maximum | 2 |
| Â | c a r d max | threshold | 10 |
| Â | Ï„ | Threshold to control the quality of latent variables | 0.3 |
|  | nb−EM−restarts | Number of random restarts for the EM algorithm | 10 |
|  | δ | Maximal physical distance (bp), to allow two SNPs in the same cluster | 50×103 |
DBSCAN | R | Maximum radius of the neighborhood to be considered to grow a cluster | value selected in 0.05 to 0.9, step 0.05 |
| Â | N min | Minimum number of points required within a cluster | 2 |