Categorical data clustering using a minimum message length criterion

Research on cluster analysis for categorical data continues to develop, new clustering algorithms being proposed. However, in this context, the determination of the number of clusters is rarely addressed. We propose a new approach in which clustering and the estimation of the number of clusters is done simultaneously for categorical data. We assume that the data originate from a finite mixture of multinomial distributions and use a minimum message length criterion (MML) to select the number of clusters (Wallace and Bolton, 1986). For this purpose, we implement an EM-type algorithm (Silvestre et al., 2008) based on the (Figueiredo and Jain, 2002) approach. The novelty of the approach rests on the integration of the model estimation and selection of the number of clusters in a single algorithm, rather than selecting this number based on a set of pre-estimated candidate models. The performance of our approach is compared with the use of Bayesian Information Criterion (BIC) (Schwarz, 1978) and Integrated Completed Likelihood (ICL) (Biernacki et al., 2000) using synthetic data. The obtained results illustrate the capacity of the proposed algorithm to attain the true number of cluster while outperforming BIC and ICL since it is faster, which is especially relevant when dealing with large data sets.

Quantity-setting competition with differentiated goods under uncertain demand

We consider a quantity-setting duopoly model, and we study the decision to move first or second, by assuming that the firms produce differentiated goods and that there is some demand uncertainty. The competitive phase consists of two periods, and in either period, the firms can make a production decision that is irreversible. As far as the firms are allowed to choose (non-cooperatively) the period they make the decision, we study the circumstances that favour sequential rather than simultaneous decisions.

Estimation of vegetation carbon stock in Portugal using land use/ land cover data

Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies

An integer programming model for the minimum interval graph completion problem

The minimum interval graph completion problem consists of, given a graph G = ( V, E ), finding a supergraph H = ( V, E ∪ F ) that is an interval graph, while adding the least number of edges |F| . We present an integer programming formulation for solving the minimum interval graph completion problem recurring to a characteri- zation of interval graphs that produces a linear ordering of the maximal cliques of the solution graph.

Constraint-based modeling of minimum set covering: application to species differentation

Work presented in the context of the European Master in Computational Logics, as partial requisit for the graduation as Master in Computational Logics

Equilibria of quantity setting differentiated duopoly with uncertainty

In this paper, we consider a Stackelberg duopoly competition with differentiated goods and with unknown costs. The firms' aim is to choose the output levels of their products according to the well-known concept of perfect Bayesian equilibrium. There is a firm ( F1 ) that chooses first the quantity 1 q of its good; the other firm ( F2 ) observes 1 q and then chooses the quantity 2 q of its good. We suppose that each firm has two different technologies, and uses one of them following a probability distribution. The use of either one or the other technology affects the unitary production cost. We show that there is exactly one perfect Bayesian equilibrium for this game. We analyse the advantages, for firms and for consumers, of using the technology with the highest production cost versus the one with the cheapest cost.

Quantity-setting competition under uncertain demand

We consider a quantity-setting duopoly model, and we study the decision to move first or second, by assuming that. the firms produce homogeneous goods and that. there is some demand uncertainty. The competitive phase consists of two periods, and in either period, the firms can make a production decision that is irreversible. As far as the firms are allowed to choose (non-cooperatively) the period they make the decision, we study the circumstances that favour sequential rather than simultaneous decisions.

Quantity competition in a differentiated duopoly

In this paper, we consider a Stackelberg duopoly competition with differentiated goods, linear and symmetric demand and with unknown costs. In our model, the two firms play a non-cooperative game with two stages: in a first stage, firm F 1 chooses the quantity, q 1, that is going to produce; in the second stage, firm F 2 observes the quantity q 1 produced by firm F 1 and chooses its own quantity q 2. Firms choose their output levels in order to maximise their profits. We suppose that each firm has two different technologies, and uses one of them following a certain probability distribution. The use of either one or the other technology affects the unitary production cost. We show that there is exactly one perfect Bayesian equilibrium for this game. We analyse the variations of the expected profits with the parameters of the model, namely with the parameters of the probability distributions, and with the parameters of the demand and differentiation.

Aircraft Engines Maintenance Costs and Reliability: An Appraisal of the Decision Process to Remove an Engine for a Shop Visit Aiming at Minimum Maintenance Unit Cost

Dissertação apresentada como requisito parcial para obtenção do grau de Mestre em Estatística e Gestão de Informação.

Minimum detection limit of an in-house nested-PCR assay for herpes simplex virus and varicella zoster virus

Introduction Herpes simplex virus (HSV) and varicella zoster virus (VZV) are responsible for a variety of human diseases, including central nervous system diseases. The use of polymerase chain reaction (PCR) techniques on cerebrospinal fluid samples has allowed the detection of viral DNA with high sensitivity and specificity. Methods Serial dilutions of quantified commercial controls of each virus were subjected to an in-house nested-PCR technique. Results The minimum detection limits for HSV and VZV were 5 and 10 copies/µL, respectively. Conclusions The detection limit of nested-PCR for HSV and VZV in this study was similar to the limits found in previous studies.

Tunnel engineering – influence of the type and the quantity of measurements in the back analysis of geomechanical parameters

The monitoring data collected during tunnel excavation can be used in inverse analysis procedures in order to identify more realistic geomechanical parameters that can increase the knowledge about the interested formations. These more realistic parameters can be used in real time to adapt the project to the real structure in situ behaviour. However, monitoring plans are normally designed for safety assessment and not especially for the purpose of inverse analysis. In fact, there is a lack of knowledge about what types and quantity of measurements are needed to succeed in identifying the parameters of interest. Also, the optimisation algorithm chosen for the identification procedure may be important for this matter. In this work, this problem is addressed using a theoretical case with which a thorough parametric study was carried out using two optimisation algorithms based on different calculation paradigms, namely a conventional gradient-based algorithm and an evolution strategy algorithm. Calculations were carried for different sets of parameters to identify several combinations of types and amount of monitoring data. The results clearly show the high importance of the available monitoring data and the chosen algorithm for the success rate of the inverse analysis process.

Influence of glucose concentration on the structure and quantity of biofilms formed by Candida parapsilosis

Candida parapsilosis is nowadays an emerging opportunistic pathogen and its increasing incidence is part related to the capacity to produce biofilm. In addition, one of the most important C. parapsilosis pathogenic risk factors includes the organisms\textquoteright selective growth capabilities in hyper alimentation solutions. Thus, in this study, we investigated the role of glucose in C. parapsilosis biofilm modulation, by studying biofilm formation, matrix composition and structure. Moreover, the expression of biofilm-related genes (BCR1, FKS1 and OLE1) were analyzed in the presence of different glucose percentages. The results demonstrated the importance of glucose in the modulation of C. parapsilosis biofilm. The concentration of glucose had direct implications on the C. parapsilosis transition of yeast cells to pseudohyphae. Additionally, it was demonstrated that biofilm related genes BCR1, FKS1 and OLE1 are involved in biofilm modulation by glucose. The mechanism by which glucose enhances biofilm formation is not fully understood, however with this study we were able to demonstrate that C. parapsilosis respond to stress conditions caused by elevated levels of glucose by up-regulating genes related to biofilm formation (BCR1, FKS1 and OLE1).

Quantity or quality? The professionalization of social media communication in Portugal and Germany: a comparison

As the worldwide growth of social media usage and institutionalization by organizations rise worldwide little is known about the degree of professionalization that has occurred. By comparing data between two asynchronous countries this research project offers insights into its strategic usage as well as discovering an interesting dynamic between activity and professionalization.

Assessing the suitability of the minimum capture size and protection regimes in the gooseneck barnacle shellfishery

The suitability of a total-length-based, minimum capture-size and different protection regimes was investigated for the gooseneck barnacle Pollicipes pollicipes shellfishery in N Spain. For this analysis, individuals that were collected from 10 sites under different fishery protection regimes (permanently open, seasonally closed, and permanently closed) were used. First, we applied a non-parametric regression model to explore the relationship between the capitulum Rostro-Tergum (RT) size and the Total Length (TL). Important heteroskedastic disturbances were detected for this relationship, demon- strating a high variability of TL with respect to RT. This result substantiates the unsuitability of a TL-based minimum size by means of a mathematical model. Due to these disturbances, an alternative growth- based minimum capture size of 26.3 mm RT (23 mm RC) was estimated using the first derivative of a Kernel-based non-parametric regression model for the relationship between RT and dry weight. For this purpose, data from the permanently protected area were used to avoid bias due to the fishery. Second, the size-frequency distribution similarity was computed using a MDS analysis for the studied sites to evaluate the effectiveness of the protection regimes. The results of this analysis indicated a positive effect of the permanent protection, while the effect of the seasonal closure was not detected. This result needs to be interpreted with caution because the current harvesting based on a potentially unsuitable mini- mum capture size may dampen the efficacy of the seasonal protection regime.

A determinação dos números de indivíduos mínimos necessários na experimentação genética

The main object of the present paper consists in giving formulas and methods which enable us to determine the minimum number of repetitions or of individuals necessary to garantee some extent the success of an experiment. The theoretical basis of all processes consists essentially in the following. Knowing the frequency of the desired p and of the non desired ovents q we may calculate the frequency of all possi- ble combinations, to be expected in n repetitions, by expanding the binomium (p-+q)n. Determining which of these combinations we want to avoid we calculate their total frequency, selecting the value of the exponent n of the binomium in such a way that this total frequency is equal or smaller than the accepted limit of precision n/pª{ 1/n1 (q/p)n + 1/(n-1)| (q/p)n-1 + 1/ 2!(n-2)| (q/p)n-2 + 1/3(n-3) (q/p)n-3... < Plim - -(1b) There does not exist an absolute limit of precision since its value depends not only upon psychological factors in our judgement, but is at the same sime a function of the number of repetitions For this reasen y have proposed (1,56) two relative values, one equal to 1-5n as the lowest value of probability and the other equal to 1-10n as the highest value of improbability, leaving between them what may be called the "region of doubt However these formulas cannot be applied in our case since this number n is just the unknown quantity. Thus we have to use, instead of the more exact values of these two formulas, the conventional limits of P.lim equal to 0,05 (Precision 5%), equal to 0,01 (Precision 1%, and to 0,001 (Precision P, 1%). The binominal formula as explained above (cf. formula 1, pg. 85), however is of rather limited applicability owing to the excessive calculus necessary, and we have thus to procure approximations as substitutes. We may use, without loss of precision, the following approximations: a) The normal or Gaussean distribution when the expected frequency p has any value between 0,1 and 0,9, and when n is at least superior to ten. b) The Poisson distribution when the expected frequecy p is smaller than 0,1. Tables V to VII show for some special cases that these approximations are very satisfactory. The praticai solution of the following problems, stated in the introduction can now be given: A) What is the minimum number of repititions necessary in order to avoid that any one of a treatments, varieties etc. may be accidentally always the best, on the best and second best, or the first, second, and third best or finally one of the n beat treatments, varieties etc. Using the first term of the binomium, we have the following equation for n: n = log Riim / log (m:) = log Riim / log.m - log a --------------(5) B) What is the minimun number of individuals necessary in 01der that a ceratin type, expected with the frequency p, may appaer at least in one, two, three or a=m+1 individuals. 1) For p between 0,1 and 0,9 and using the Gaussean approximation we have: on - ó. p (1-p) n - a -1.m b= δ. 1-p /p e c = m/p } -------------------(7) n = b + b² + 4 c/ 2 n´ = 1/p n cor = n + n' ---------- (8) We have to use the correction n' when p has a value between 0,25 and 0,75. The greek letters delta represents in the present esse the unilateral limits of the Gaussean distribution for the three conventional limits of precision : 1,64; 2,33; and 3,09 respectively. h we are only interested in having at least one individual, and m becomes equal to zero, the formula reduces to : c= m/p o para a = 1 a = { b + b²}² = b² = δ2 1- p /p }-----------------(9) n = 1/p n (cor) = n + n´ 2) If p is smaller than 0,1 we may use table 1 in order to find the mean m of a Poisson distribution and determine. n = m: p C) Which is the minimun number of individuals necessary for distinguishing two frequencies p1 and p2? 1) When pl and p2 are values between 0,1 and 0,9 we have: n = { δ p1 ( 1-pi) + p2) / p2 (1 - p2) n= 1/p1-p2 }------------ (13) n (cor) We have again to use the unilateral limits of the Gaussean distribution. The correction n' should be used if at least one of the valors pl or p2 has a value between 0,25 and 0,75. A more complicated formula may be used in cases where whe want to increase the precision : n (p1 - p2) δ { p1 (1- p2 ) / n= m δ = δ p1 ( 1 - p1) + p2 ( 1 - p2) c= m / p1 - p2 n = { b2 + 4 4 c }2 }--------- (14) n = 1/ p1 - p2 2) When both pl and p2 are smaller than 0,1 we determine the quocient (pl-r-p2) and procure the corresponding number m2 of a Poisson distribution in table 2. The value n is found by the equation : n = mg /p2 ------------- (15) D) What is the minimun number necessary for distinguishing three or more frequencies, p2 p1 p3. If the frequecies pl p2 p3 are values between 0,1 e 0,9 we have to solve the individual equations and sue the higest value of n thus determined : n 1.2 = {δ p1 (1 - p1) / p1 - p2 }² = Fiim n 1.2 = { δ p1 ( 1 - p1) + p1 ( 1 - p1) }² } -- (16) Delta represents now the bilateral limits of the : Gaussean distrioution : 1,96-2,58-3,29. 2) No table was prepared for the relatively rare cases of a comparison of threes or more frequencies below 0,1 and in such cases extremely high numbers would be required. E) A process is given which serves to solve two problemr of informatory nature : a) if a special type appears in n individuals with a frequency p(obs), what may be the corresponding ideal value of p(esp), or; b) if we study samples of n in diviuals and expect a certain type with a frequency p(esp) what may be the extreme limits of p(obs) in individual farmlies ? I.) If we are dealing with values between 0,1 and 0,9 we may use table 3. To solve the first question we select the respective horizontal line for p(obs) and determine which column corresponds to our value of n and find the respective value of p(esp) by interpolating between columns. In order to solve the second problem we start with the respective column for p(esp) and find the horizontal line for the given value of n either diretly or by approximation and by interpolation. 2) For frequencies smaller than 0,1 we have to use table 4 and transform the fractions p(esp) and p(obs) in numbers of Poisson series by multiplication with n. Tn order to solve the first broblem, we verify in which line the lower Poisson limit is equal to m(obs) and transform the corresponding value of m into frequecy p(esp) by dividing through n. The observed frequency may thus be a chance deviate of any value between 0,0... and the values given by dividing the value of m in the table by n. In the second case we transform first the expectation p(esp) into a value of m and procure in the horizontal line, corresponding to m(esp) the extreme values om m which than must be transformed, by dividing through n into values of p(obs). F) Partial and progressive tests may be recomended in all cases where there is lack of material or where the loss of time is less importent than the cost of large scale experiments since in many cases the minimun number necessary to garantee the results within the limits of precision is rather large. One should not forget that the minimun number really represents at the same time a maximun number, necessary only if one takes into consideration essentially the disfavorable variations, but smaller numbers may frequently already satisfactory results. For instance, by definition, we know that a frequecy of p means that we expect one individual in every total o(f1-p). If there were no chance variations, this number (1- p) will be suficient. and if there were favorable variations a smaller number still may yield one individual of the desired type. r.nus trusting to luck, one may start the experiment with numbers, smaller than the minimun calculated according to the formulas given above, and increase the total untill the desired result is obtained and this may well b ebefore the "minimum number" is reached. Some concrete examples of this partial or progressive procedure are given from our genetical experiments with maize.