Screen Searching Sequential

Over the past several years work has been underway on the second generation parallel structure search architecture. This architecture is a completely new hardware and software implementation which builds on the already proven architectural principles of the first generation system as well as taking advantage of inverted file searching rather than sequential searching to implement the screen search algorithm. 

In response to these requirements, a structure search system was designed and by mid-1980 a pilot system was built and being tested. The pilot system had only a subset of the features that users would require and it searched only about 5 per cent of the entire CAS Registry structure file. However, it did demonstrate the feasibility of using serial sequential screen searching which was a critical design concept of the system. 

Screen searching is done serially rather than using a more traditional inverted file approach. That is, the entire screen file for a particular file segment is read sequentially from disk into memory. All sets of query screens on the current list of queries to be searched are compared against each file structure screen set. If a query screen set matches a file screen set, then the candidate file structure is passed on to iterative search. 

This sequential screen search had the following advantages ... 

The search engine architecture will substitute inverted file searching for sequential screen searching. This will permit the structure screen set size to be increased because new screens simply become new terms in the inverted dictionary. It will also permit other numeric or textual information associated with a structure to be added to the inverted screen file and processed as part of the inverted screen searching process. 

For these reasons we turned first to a version of seriation that had no specific spatial task requirement only a search one. Using computer driven touch screens, we displayed a variety of icons varying in size in random linear arrays, and required that subjects touch each one once only, according to a sequential rule such as biggest to smallest. The randomized arrays ensured that there was no correlation between the way the sizes appeared, and (e.g.) the monotonic search path that had to be complied with (see Figure 10.5a). In this way, we have been able to assess size seriation, considered as an extended serial search task without manipulation requirements and spatial confounds. 

The above considerations concern atom-by-atom substructure searches, which in practice are only the final steps of a substructure query handling cycle since, usually, some screening steps are utilised to substantially narrow down the number of potential hits. Only the relatively few hit candidates are then subject to atom-by-atom searches. It is worth considering the effects of such screening on the expected comparative performance of sequentially performed versus h3q>erstructure based (i.e., quasiparallel) procedures. 

The CSD is released, twice each year, as an indexed sequential binary file that contains additional data items derived from the basic input. Principally, these are bit-encoded screens that are required to enhance search speeds within the CSD software. Most screens are derived from the 2D connectivity... 

The screen index file is the first file accessed for any structure or sula-structure search. It contains all the information necessary to determine structure matches. When the structure matches have been located, additional information for each structure, such as the structure picture, may be retrieved via the accession number. Because its orgcuiization is index-sequential, the screen index file may iye accessed either sequentially, or selectively by use of the its Indexes. 

---