Yung-Hsing Peng

Genetic algorithms for the investment of the mutual fund with global trend indicator

Our investment strategy for the world mutual funds can be divided into three main parts. First, the global trend indicator (GTI) is defined for evaluating the price change trend of the funds in the future. Then, based on GTI, we derive the monitoring indicator (MI) to measure whether the fund market is in the bull or bear state. Finally, to decide the signal for buying or selling funds, a genetic algorithm is invoked to dynamically select funds according to their past performances (profitability). In our experimental results from January 1999 to December 2008 (10years in total), we achieve the annual profit higher than 10%.

AN ALGORITHM AND APPLICATIONS TO SEQUENCE ALIGNMENT WITH WEIGHTED CONSTRAINTS

Given two sequences S1, S2, and a constrained sequence C, a longest common subsequence of S1, S2 with restriction to C is called a constrained longest common subsequence of S1 and S2 with C. At the same time, an optimal alignment of S1, S2 with restriction to C is called a constrained pairwise sequence alignment of S1 and S2 with C. Previous algorithms have shown that the constrained longest common subsequence problem is a special case of the constrained pairwise sequence alignment problem, and that both of them can be solved in O(rnm) time, where r, n, and m represent the lengths of C, S1, and S2, respectively. In this paper, we extend the definition of constrained pairwise sequence alignment to a more flexible version, called weighted constrained pairwise sequence alignment, in which some constraints might be ignored. We first give an O(rnm)-time algorithm for solving the weighted constrained pairwise sequence alignment problem, then show that our extension can be adopted to solve some constraint-related problems that cannot be solved by previous algorithms for the constrained longest common subsequence problem or the constrained pairwise sequence alignment problem. Therefore, in contrast to previous results, our extension is a new and suitable model for sequence analysis.

Efficient algorithms for finding interleaving relationship between sequences

The longest common subsequence and sequence alignment problems have been studied extensively and they can be regarded as the relationship measurement between sequences. However, most of them treat sequences evenly or consider only two sequences. Recently, with the rise of whole-genome duplication research, the doubly conserved synteny relationship among three sequences should be considered. It is a brand new model to find a merging way for understanding the interleaving relationship of sequences. Here, we define the merged LCS problem for measuring the interleaving relationship among three sequences. An O(n^3) algorithm is first proposed for solving the problem, where n is the sequence length. We further discuss the variant version of this problem with the block information. For the blocked merged LCS problem, we propose an algorithm with time complexity O(n^2m), where m is the number of blocks. An improved O(n^2+nm^2) algorithm is further proposed for the same blocked problem.

Dynamic programming algorithms for the mosaic longest common subsequence problem

The longest common subsequence (LCS) problem can be used to measure the relationship between sequences. In general, the inputs of the LCS problem are two sequences. For finding the relationship between one sequence and a set of sequences, we cannot apply the traditional LCS algorithms immediately. In this paper, we define the mosaic LCS (MLCS) problem of finding a mosaic sequence C, composed of repeatable k sequences in source sequence set S, such that the LCS of C and the target sequence T is maximal. Based on the concept of break points in sequence T, we first propose a divide-and-conquer algorithm with O(n^2m|S|+n^3logk) time for solving this problem, where n and m are the length of T and the maximal length of sequences in S, respectively. Furthermore, an improved algorithm with O(n(m+k)|S|) time is proposed by applying an efficient preprocessing for the MLCS problem.

Efficient algorithms for the block edit problems

In this paper, we focus on the edit distance between two given strings where block-edit operations are allowed and better fitting to the human natural edit behaviors. Previous results showed that this problem is NP-hard when block moves are allowed. Various approximations to this problem have been proposed in recent years. However, this problem can be solved by the polynomial-time optimization algorithms if some reasonable restrictions are applied. The restricted variations which we consider involve character insertions, character deletions, block copies and block deletions. In this paper, three problems are defined with different measuring functions, which are P(EIS,C), P(EI,L) and P(EI,N). Then we show that with some preprocessing, the minimum block edit distances of these three problems can be obtained by dynamic programming in O(nm), O(nmlogm) and O(nm^2) time, respectively, where n and m are the lengths of the two input strings.

The trading on the mutual funds by gene expression programming with Sortino ratio

The aim of this paper is to combine several techniques together to provide one systematic method for guiding the investment in mutual funds. Many researches focus on the prediction of a single asset time series, or focus on portfolio management to diversify the investment risk, but they do not generate explicit trading rules. Only a few researches combine these two concepts together, but they adjust trading rules manually. Our method combines the techniques for generating observable and profitable trading rules, managing portfolio and allocating capital. First, the buying timing and selling timing are decided by the trading rules generated by gene expression programming. The trading rules are suitable for the constantly changing market. Second, the funds with higher Sortino ratios are selected into the portfolio. Third, there are two models for capital allocation, one allocates the capital equally (EQ) and the other allocates the capital with the mean variance (MV) model. Also, we perform superior predictive ability test to ensure that our method can earn positive returns without data snooping. To evaluate the return performance of our method, we simulate the investment on mutual funds from January 1999 to September 2012. The training duration is from 1999/1/1 to 2003/12/31, while the testing duration is from 2004/1/1 to 2012/9/11. The best annualized return of our method with EQ and MV capital allocation models are 12.08% and 12.85%, respectively. The latter also lowers the investment risk. To compare with the method proposed by Tsai et al., we also perform testing from January 2004 to December 2008. The experimental results show that our method can earn annualized return 9.07% and 11.27%, which are better than the annualized return 6.89% of Tsai et al.

An effective wavelength utilization for spectroscopic analysis on orchid chlorophyll measurement

To insure the quality and quantity of yields, computational tools for monitoring and analyzing the growth of crops are of great importance in scientific agriculture. In recent years, non-destructive measurements that utilize spectroscopy for crop monitoring have drawn much attention, and algorithms for selecting proper wavelengths are worth being investigated, since they have deep impact on the accuracy. In this research, an approach for utilizing wavelengths on orchid chlorophyll prediction is proposed. The newly proposed method is based on the response surface methodology (RSM), and we apply it to four wavelength selection algorithms to see the effectiveness. The spectral data in our experiment is obtained by the interactance measurement on 600 orchid plants with a hand-held spectrometer, and the actual chlorophyll content is also measured with a CCI meter for verification. Experimental results show that this new approach significantly improves the utilization of wavelengths for building prediction model, raising R2 from 88.74% to 93.95% and reducing the RMSECV from 7.5 to 6.94 for 15 wavelengths. Therefore, the proposed method is worth being applied to devising wavelength selection algorithms.

An investigation of spacial approaches for crop price forecasting in different Taiwan markets

In agri-business, the market prices of crops are indicators for their demands and supplies. Therefore, crop price forecasting is a research of high interest, which can be used to detect potential selling points to increase the profit. For production management, these selling points can be deemed as recommended harvest time, and better cultivation schedules could be derived accordingly. In Taiwan, the Council of Agriculture (COA) establishes an official website that publishes crop prices from 19 local markets, which covers more than 100 different crops. This website is now widely referred by farmers and agri-enterprises as a channel to keep track of market information. Therefore, the published daily prices are good materials for research of data science. In this paper, we aim to investigate the spacial relationships between prices in different markets. The investigation is accomplished by examining the forecasting price given by four well-known spacial algorithms, which are the nearest neighbor (NN), the inverse distance weighting (IDW), the Kriging method, and the artificial neural network (ANN). We compare the performance of these four algorithms with the price data obtained from 15 markets on the official website of COA. The experimented crops are cabbage, bok choy, watermelon, and cauliflower. According to the experimental results, Kriging achieves the lowest error in percentages. Considering the time efficiency, the Kriging method is also recommended for the development of forecasting service, since the regression can be accomplished efficiently.

Developing crop price forecasting service using open data from Taiwan markets

From the perspective of agricultural business, the market price of certain crop reflects the demand of that crop in current stage. Therefore, to track and to forecast the market prices are both important tasks in agri-management, by which the production schedule can be adjusted to increase the profit. For tracking the crop prices, the Council of Agriculture (COA) establishes an official website that provides open data of daily market prices from over 15 local markets with more than 100 different crops in Taiwan. Recently, the smart agri-management platform (S.A.M.P.) is developed by the Institute for Information Industry (III) as an integrated cloud service for agri-business. Inspired by the open data of crop prices, in this paper we develop a crop price forecasting service on S.A.M.P., which automatically retrieves the historical prices on the official website as training dataset, and provides the price forecasting service with some well-known algorithms for time series analysis. The algorithms implemented in this paper are the autoregressive integrated moving average (ARIMA), the partial least square (PLS), and the artificial neural network (ANN). In addition, for PLS we further integrate the response surface methodology (RSM), deriving a new algorithm RSMPLS, by which the non-linear relationship between historical prices can be investigated. We compare the performance of these four algorithms with the price data obtained from the First Fruit and Vegetable Wholesale Market in Taipei. The experimented crops are cabbage, bok choy, watermelon, and cauliflower. According to the experimental results, PLS and ANN are of lower error in percentages. In addition, PLS and ANN are recommended for short term and long term forecasting, respectively.

An Efficient RSM-Based Algorithm for Measuring Chlorophyll on Orchid Leaves with a Microspectrometer

As the manufacturing and the computing power of mobile devices improve, micro-instruments have more and more applications nowadays. In the field of agricultural science, a spectrometer provides a non-destructive way to extract the inner growth feature of plants, which helps to keep track of the health of crops, and helps to determine the required treatment. In this paper, we propose an algorithm that predicts the concentration of chlorophyll on orchid leaves, by analyzing the spectral data collected from leaves with a micro spectrometer. The proposed algorithm utilizes the response surface methodology for building a non-linear prediction model. For verifying our algorithm, we collect 400 spectral data from four different species of orchids, where each individual species contains 100 samples. The experimental results show that our prediction model using 8 different wavelengths achieves 0.842 and 9.14 in R2 and RMSECV, respectively, which is competitive to the result of traditional approach using 43 different wavelengths.