Data Partitioning

Step 1. From a histogram of the data, partition the data into N components, each roughly corresponding to a mode of the data distribution. This defines the Cj. Set the parameters for prior distributions on the 6 parameters that are conjugate to the likelihoods. For the normal distribution the priors are defined in Eq. (15), so the full prior for the n components is... 

Figure 2.6 Percentiles for solubility for training set fragments as a function of mean Clog P for each bin. Data partitioned by Clog P into 10 bins each with 356 solubility measurements. Reprinted with permission from [23].

The well-known domain decomposition (DD) methods, see e.g. Chan Mathew (1994), use data partition induced by a decomposition of the computational domain. This decomposition can be used for two purposes Firstly for parallel implementation of vector updates, inner products and matrix-vector multiplication, i.e. for parallel implementation of the CG method. Secondly, for a construction of efficient preconditioners. 

Table 1. Brain penetration data, partition coefficients and histamine H receptor antagonist activities for selected H2 antagonists. Data from Young et al. [24].

Nonsmooth response surface for the fake data partitioning as in Figure 5.34. 

Type of data partition coefficient at equilibrium swelling, = Cq (in swollen PDMS)/cc (in CO2)... 

The development of algorithms that look for local patterns in the data—data partitioning methods have proved to be a promising approach... 

Domain and co-domain splitting perform a sort of functional partitioning while data partitioning is introduced to fully exploit the whole memory available in a parallel machine. 

Define tasks to be performed on each data partition. 

On selected cluster, allocate nodes to data partitions. 

Table 2.2 indicates free congruence in one-gene data partitions compared to the combined evidence tree (Figure 2.1 frees from gene partitions not shown). Combined analysis of all genes results in six clades that are not produced in consensus MPTs from one-gene partitions. The consensus MPTs... 

Architectural Partitioning none none none none 2 data partitions... 

Partitioning does not have a large effect on the scheduling of this modular design. The partitions indicate that a PLUS operation may execute in the address partition at the same time that another PLUS operation is executing in the data partition, but it is never necessary to schedule simultaneous PLUS operations, so the partitions do not alter the schedule. 

The address partition is shown at the top of Figure 8-6 and the data partition is shown at the bottom. The sequencer and the small memories are not considered to be part of any partition, while memories ANEW and K each make up partitions of their own. The quality of this partitioning scheme for the Kalman Filter can be seen in the interconnect of the data path. There are no direct connections between tile address partition (at the top of the figure) and the data partition (at the bottom of the figure). All connections are via the memories and busses that sit between the two partitions. 

The data partition requires an ALU, a shifter, and a comparator. The ALU and shifter are separate functional units because they are of different bitwidths. The comparator is separated from the ALU due to the functional costs assumed for this design. The functional costs for these operations were specified to be high so that comparison and ALU functions could execute in the same cstep, as is done in the commercial design. A difference between the general path design and the commercial design is that the commercial design implements the ALU... 

---