Study of Inputs and Outputs. Every algorithm has a set of inputs that are needed to be calculated to produce a set of outputs. Inputs and Outputs define the interface between the computational system and the user. Inputs study consists of the data volume, their width, their nature, their source, and if there are used more than one time in the calculations. Volume could be the most important parameter for the study of the algorithm as it can lead to very important decisions, such as use of memory, number of processing units etc. If input data is very little then the system does not need any special handling for them as internal resources can handle them. If input data becomes bigger, and internal memory is not enough for handling their volume, then internal structures should be designed in order to compress or encode data, and, finally if volume of data is big, that reconfigurable device cannot store them, then an system e.g. external memory based system for buffering or swapping, should be designed to manipulate them. Data size is little or big depending on the target reconfigurable technology and device which is used from the designer. The size and the characteristics of the input are important as for example if it is very large, and comes through conventional protocols (Ethernet, PCI etc) then special data structures should be designed to reassemble data. Such structures can be shift registers, very long word registers etc., which handle input data and reform them to the appropriate format for reconfigurable hardware processing. The input data rate is also an important aspect to a system design. If the algorithm handles input data in burst mode data structures as FIFOs should be designed. If there is input data that an algorithm uses more than one time then the corresponding data structures, as cashing schemes, cyclic buffers, local memories, should also be used. It is also important, for the designer, to study the if there is a need to design the controller for the protocol that the inputs may follow. If this system, for example, is directly connected to network with TCP/IP for streaming inputs, or an external DDR memory for large databases etc. The proper controller in hardware should be designed and integrated to the system. The designer handles outputs similarly to the inputs. For QualiMaster scope Input and Output study is critical as it handles streaming data. Inputs and outputs have been often proved to be a critical issue to designers for reconfigurable hardware, constrainin...
Study of Inputs and Outputs. The Count-Min sketch provides a different kind of solution to count tracking. It uses a fixed amount of memory to store count information, which does not vary over time. Nevertheless, it is able to provide useful estimated counts, as the accuracy scales with the total sum of all the counts stored. Streaming data from different applications, like IP networking, machine learning, distributed computing and signal processing, can be used as input to the proposed algorithm. The Count Min sketch is a data structure that summarizes efficiently a stream of input data and answers queries with high accuracy over the input stream. Thus, the CM algorithm takes two different types of inputs: a stream of input data that is summed up on the CM sketch data structure or a query over the input dataset.
Study of Inputs and Outputs. The EH model offers a probabilistic solution to the counting problem. The EH algorithm either updates the Exponential Histogram data structure with new streaming data or it estimates the number of 1s that have arrived from a specific timestamp up to the current timestamp.
Study of Inputs and Outputs. As described above, the Xxxxxxx-Xxxxxxx (HY) Correlation Estimator measures the pair wise correlation of the input market stocks. The HY Correlation Estimator uses the transaction prices of two stocks in order to calculate their correlation. The correlation is calculated over time intervals that the stocks transactions take place. On the other hand, the QualiMaster project focuses on the correlation among a group of market stocks. The HY estimator calculates the correlation of all the different pairs of the processing market stocks. Figure 11 presents the equation for calculating the HY estimator for two different market stocks.
Study of Inputs and Outputs. The SVM Training algorithm receives as input a two dimensional structure and outputs the SVM model.
Study of Inputs and Outputs. LDA takes as input a set of documents and posits that each document is a mixture of a small number of topics and that each word's creation is attributable to one of the document's topics. The applications that implement LDA take as input a filtered set of documents, from stop words, common used words etc., or filters a set of documents. They then create an index table with the vocabulary and use the indexes in processing. The training step of the algorithm produces a number of topics (input by user). The basic idea is that documents are represented as random mixtures over latent topics, where each topic is characterized by a distribution over words. The model produced by the training is then used to characterize new documents. Each new document is characterized by a different set of topics.
