Distributed Progressive Sequential Pattern Mining on the Cloud
- 格式:pdf
- 大小:301.23 KB
- 文档页数:8
IJG(Independent JPEG Group,联合图像专家组)IJG JPEG库:系统的体系结构。
本文件是JPEG软件的一部分。
使用和销售的条件,请参照随带的readme文件。
本文件给出了JPEG软件的大体结构,包括系统中各个模块中的函数及模块间的接口。
更多关于数据结构和行业协议的详解请参照包含文件和源程序中的注释。
.我们假设读者已经对JPEG标准有了一定的了解。
Readme文件中列出了关于的JPEG 的参考书目。
文件libjpeg.doc从程序师实际应用的角度描述了库,因此最好先读那个文件再来看本文件。
同时,coderules.doc文件介绍了代码中的类型规定。
本文件中,JPEG的专业术语遵循JPEG标准。
“component”:颜色信道。
例如:红色或亮度。
“sample”:采样单元。
(某一图像数据的数)。
“coefficient”:频率系数。
(DCT变换的输出数)。
“block”:采样单元或频率系数中的一个8*8的组。
“MUC”(最小编码单元):隔行扫描机制中块的大小,由采样因数决定;或某一非隔行扫描中的单独块。
我们不交叉使用术语“pixel”和“sample”。
当说到“pixel”时他代表全幅图像的一个成分,而“sample”指采样图像的一成分。
因此“sample”的数量可能经颜色信道发生改变而“pixel”不会改变。
(这一术语区分没有严格的贯穿代码始终,但使用在那些一旦混淆就将导致错误的地方。
)***系统特征***IJG的发行包括两部分:* JPEG压缩与解压的子程序库。
* cjpeg/djpeg,两个应用库,转化JFIF、JPEG到其他图像格式的软件实例。
cjpeg/djpeg不是很复杂,他们仅仅加入了对几种未压缩的图像格式使用接口和I/O程序的简单命令行。
这个文档浓缩在库本身中。
我们希望这个库能够支持所有的JPEG基线,甚至顺序和向前DCT处理。
但不支持分级处理。
The library does not support the lossless (spatial) JPEG process. Lossless JPEG shares little or no code with lossy JPEG, and would normally be used without the extensive pre- and post-processing provided by this library.We feel that lossless JPEG is better handled by a separate library.Within these limits, any set of compression parameters allowed by the JPEG spec should be readable for decompression. (We can be more restrictive about what formats we can generate.) Although the system design allows for all parameter values, some uncommon settings are not yet implemented and maynever be; nonintegral sampling ratios are the prime example. Furthermore,we treat 8-bit vs. 12-bit data precision as a compile-time switch, not arun-time option, because most machines can store 8-bit pixels much more compactly than 12-bit.For legal reasons, JPEG arithmetic coding is not currently supported, but extending the library to include it would be straightforward.By itself, the library handles only interchange JPEG datastreams --- in particular the widely used JFIF file format. The library can be used by surrounding code to process interchange or abbreviated JPEG datastreams that are embedded in more complex file formats. (For example, libtiff uses this library to implement JPEG compression within the TIFF file format.)The library includes a substantial amount of code that is not covered by the JPEG standard but is necessary for typical applications of JPEG. These functions preprocess the image before JPEG compression or postprocess it after decompression. They include colorspace conversion, downsampling/upsampling, and color quantization. This code can be omitted if not needed.A wide range of quality vs. speed tradeoffs are possible in JPEG processing, and even more so in decompression postprocessing. The decompression library provides multiple implementations that cover most of the useful tradeoffs, ranging from very-high-quality down to fast-preview operation. On thecompression side we have generally not provided low-quality choices, since compression is normally less time-critical. It should be understood that the low-quality modes may not meet the JPEG standard's accuracy requirements; nonetheless, they are useful for viewers.*** Portability issues ***Portability is an essential requirement for the library. The key portability issues that show up at the level of system architecture are:1. Memory usage. We want the code to be able to run on PC-class machineswith limited memory. Images should therefore be processed sequentially (in strips), to avoid holding the whole image in memory at once. Where afull-image buffer is necessary, we should be able to use either virtual memory or temporary files.2. Near/far pointer distinction. To run efficiently on 80x86 machines, the code should distinguish "small" objects (kept in near data space) from "large" ones (kept in far data space). This is an annoying restriction, but fortunately it does not impact code quality for less brain-damaged machines,and the source code clutter turns out to be minimal with sufficient use of pointer typedefs.3. Data precision. We assume that "char" is at least 8 bits, "short" and "int" at least 16, "long" at least 32. The code will work fine with largerdata sizes, although memory may be used inefficiently in some cases. However, the JPEG compressed datastream must ultimately appear on external storage as a sequence of 8-bit bytes if it is to conform to the standard. This may pose a problem on machines where char is wider than 8 bits. The library representscompressed data as an array of values of typedef JOCTET. If no data type exactly 8 bits wide is available, custom data source and data destination modules must be written to unpack and pack the chosen JOCTET datatype into8-bit external representation.*** System overview ***The compressor and decompressor are each divided into two main sections:the JPEG compressor or decompressor proper, and the preprocessing or postprocessing functions. The interface between these two sections is the image data that the official JPEG spec regards as its input or output: this data is in the colorspace to be used for compression, and it is downsampledto the sampling factors to be used. The preprocessing and postprocessingsteps are responsible for converting a normal image representation to or from this form. (Those few applications that want to deal with YCbCr downsampled data can skip the preprocessing or postprocessing step.)Looking more closely, the compressor library contains the following main elements:Preprocessing:* Color space conversion (e.g., RGB to YCbCr).颜色空间变换* Edge expansion and downsampling. Optionally, this step can do simplesmoothing --- this is often helpful for low-quality source data.边缘扩展和降低采样,这个可选项使采样平滑。
写循序渐进的作文开头英文回答:Starting a composition can sometimes be a challenging task. It requires creativity and a clear direction to captivate the readers' attention. In my opinion, there are a few effective ways to begin an essay in a sequential and progressive manner.One approach is to start with a thought-provoking question. For instance, "Have you ever wondered why some people are more successful than others?" This immediately engages the readers and encourages them to think about the topic. By posing a question, I can introduce the main idea of my essay while piquing the readers' curiosity.Another effective way to start an essay is by sharing a personal anecdote or story. By narrating a relatable experience, I can establish a connection with the readers. For example, "I vividly remember the day when I received myfirst acceptance letter from university. It was a moment of triumph and relief, but it also made me reflect on the factors that contributed to my success." Sharing personal stories adds a human touch to the essay and makes it more relatable.Additionally, using a relevant quote or a famous saying can be an impactful way to begin an essay. For instance, "As the saying goes, 'Success is not the key to happiness. Happiness is the key to success.' This quote by Albert Schweitzer perfectly encapsulates the importance of finding joy in what we do." By incorporating a quote, I can set the tone for my essay and provide a perspective that supports my argument.中文回答:写作文的开头有时候是一项具有挑战性的任务。
电视原理单词表部门: xxx时间: xxx整理范文,仅供参考,可下载自行编辑Chapter OneAFC Automatic Frequency Controlaperture 光圈,孔, 穴, 缝隙A/V Audio & Videobaseband 基带blanking 消隐vertical ~场消隐horizontal ~行消隐branch line 支线CCD charge-coupled device 电耦合器件chroma 色度~keying 色键chrominance 色度closed-circuit television(CATV> 闭路电视coaxial cable 同轴电缆color bar 彩条convergence 聚焦costing 涂敷crystal 晶体CRT cathode-ray tube 阴极射线管distortion 畸变DSP Digital Signal Processing EIA Electronic Industries Association 电子工业协会even 偶数facsimile FAX 传真FCC Federal Communications Commission (美国>通信委员会flexible cable 柔性电缆fluorescent 荧光的,~screen 荧光屏flyback 回扫FSK Frequency-Shift Keying 频率键控geosynchronous 与地球的相对位置不变的,同步的handshake 握手信号HDTV High-Definition TV 高清晰度电视Heterodyned 外差的home theater 家庭影院horizontal 水平的independence 阻抗Infrared 红外线interlaced scan 隔行扫描iris 光圈jitter 抖动laserdisc 激光视盘light-of-sight 视线luminance 亮度modulation 调制monochrome 单色nonsegmented 非分割的NTSC National Television System Committee 国家电视制式委员会b5E2RGbCAPoctave 倍频程,阶程,八度音阶odd 奇数oxide 氧化的pan 全景旋转picture-in-a-picture 画中画progressive scan 逐行扫描projection 投射propagate 传播raster 光栅rear projection 背投front projection 直投record-playback 记录回放redundancy 冗余refract vt.使折射, 测定...的折射度reflect v.反射, 反映, 表现, 反省, 细想retrace 折回,回扫,逆程,field ~场回扫scrambled 混杂的signal compression 信号压缩spatial adj.空间的special-effects 特殊效果splice 连接,合并stereo 立体声subscriber 用户subcarrier 副载波surround sound 环绕声surveillance 监视sweep 扫描temporal adj.时间的, 当时的, 暂时的, 现世的, 世俗的, [解]颞的time base 时基tilt 倾斜time lapse 慢速拍摄tri-chromatic adj.三色的, 三色版的, 三原色的trunk line 主干线VCR videocassette recorder 卡式录像机vertical 垂直的VHF very high-frequency 超高频vidicon 光导摄像管voiceover 话外音UHF 甚高频Chapter Twoaural 听觉的bias 偏置,偏压brightness 亮度contrast 对比度deflection 偏转duplicate 复制flutter 颤动flyback time 返回时间frame 帧ghost 重影half wave 半波full wave 全波hue 色调intercarrier 间载波luminance 亮度overlay 覆盖peak-to-peak 峰-峰值persistence 持续性,余辉persistence of vision 视觉后滞,视觉滞留photoengrave 照相雕刻photograph 像片pixel 像素polarity 极性up ~, positive ~正极性resolution 分辨率ripple 纹波saturation 饱和度sequential 连续的,顺序的shutter 快门sideband 边带solid-state 固态superimposed 成阶层的, 有层理的tint 色调transducer 转换器VSB vestigial sideband 残留边带vestigial 残留的Chapter Threeauto-focus 自动聚焦burst 脉冲,爆发calibrate 校准camcorder 摄录机candela 烛光candle 烛光candlepower 烛光chalnicon 硒化镉光导摄像管channel stopper 沟道截断环clamp 钳位,夹住columnar 分纵栏的,柱形的, 筒形的composite 合成的;合成物console 控制台cyan 蓝绿色, 青色decode 解码encode 编码depth of field 景深dichroic 二向色的, 二色性的dictate 指示, 命令, 规定dissector 解剖器,折像管exponent 指数f-stop number the ratio of focal length to thediameter of thelenp1EanqFDPwfilter 滤波器flowchart 流程图flux 通量light ~光通量flying dot scanner 飞点扫描器focal length 焦距footcandle 尺烛光(照度单位> frame transfer FT 帧转移gamma correction γ校正glitch 短时脉冲波形干扰graticule 分成小方格halogen[化] 卤素~卤素灯iconoscope 光电摄像管intercom 对讲电话装置illumination 光照强度incandescent 白炽的logarithm 对数lux 勒克斯<照明单位)macro 宏观的,大的nanometer 十亿分之一公尺,毫微Mm μnewvicon 碲化锌镉视像管nonlinear 非线性obsolete 荒废的, 陈旧的orthicon 正折像管pedestal .基准,基础photoconductive 光导photodiode 光二极管photoelectric 光电的photoemissive 光发射photosensitive 感光的plumbicon 氧化铅摄像管preamplifier 预放大器prism 棱镜retention 余辉rounding upward 向上取整,四舍五入saticon 硒砷碲视像管scattering 散射shift register 位移寄存器spurious 伪造的, 假造的, 欺骗的sturdy 坚固的synchronize 同步taking lens 取像透镜, 拍摄镜头telephoto 用远距镜头照相的transistor 晶体管troubleshooting 故障检修,故障查找vector 矢量~scope 矢量显示器viewfinder 取景器wide-angle 广角zoom 变焦Chapter Fourafterglow 余辉alloy 合金amber 琥珀ambient 周围的,周围环境anode 阳极aquadag 导电敷层attenuate 衰减backing 基础。
Journal of Experimental Botany, Vol. 65, No. 4, pp. 1193–1203, 2014doi:10.1093/jxb/ert482 Advance Access publication 24 January, 2014This paper is available online free of all access charges (see /open_access.html for further details)ReseaRch papeRSequential action of FRUITFULL as a modulator of the activity of the floral regulators SVP and SOC1Vicente Balanzà, Irene Martínez-Fernández and Cristina Ferrándiz*Instituto de Biología Molecular y Celular de Plantas, Consejo Superior de Investigaciones Científicas–Universidad Politécnica de Valencia, Avenida de los Naranjos s/n, 46022 Valencia, Spain*To whom correspondence should be addressed. E-mail: cferrandiz@ibmcp.upv.esReceived 18 July 2013; Revised 11 November 2013; Accepted 12 December 2013AbstractThe role in flowering time of the MADS-box transcription factor FRUITFULL (FUL) has been proposed in many works. FUL has been connected to several flowering pathways as a target of the photoperiod, ambient temperature, and age pathways and it is has been shown to promote flowering in a partially redundant manner with SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (SOC1). However, the position of FUL in these genetic networks, as well as the functional output of FUL activity during floral transition, remains unclear. In this work, a genetic approach has been undertaken to understand better the functional hierarchies involving FUL and other MADS-box factors with well established roles as floral integrators such as SOC1, SHORT VEGETATIVE PHASE (SVP) or FLOWERING LOCUS C (FLC). Our results suggest a prominent role of FUL in promoting reproductive transition when photoinductive signal-ling is suppressed by short-day conditions or by high levels of FLC expression, as in non-vernalized winter ecotypes.A model is proposed where the sequential formation of FUL–SVP and FUL–SOC1 heterodimers may mediate the veg-etative and meristem identity transitions, counteracting the repressive effect of FLC and SVP on flowering.Key words:Flowering, FUL, SVP, SOC1, FLC, MADS-box factors.IntroductionArabidopsis thaliana adult life cycle comprises three major phase transitions that are mainly characterized by the identity of the lateral structures produced by the shoot apical meris-tem (SAM). The vegetative phase transition marks the change from the production of juvenile leaves to the production of adult leaves. Both types of leaves form a rosette through the period of vegetative growth of the plant and, then, trig-gered by both environmental and endogenous cues, the SAM undergoes two subsequent phase transitions leading to repro-ductive development: the reproductive transition that causes bolting of the primary inflorescence and the production of cauline leaves subtending secondary inflorescences, and the meristem identity transition, after which the SAM will pro-duce floral meristems directly (Araki, 2001; Yamaguchi et al., 2009; Huijser and Schmid, 2011).Both reproductive and meristem identity transitions, that are collectively named as floral transition, are highly controlled by developmental and environmental signals. Six promoting pathways have been proposed to regulate this pro-cess (reviewed in Fornara et al., 2010; Srikanth and Schmid, 2011): the photoperiod, vernalization, ambient temperature, age, autonomous, and gibberellin pathways. The first three pathways respond to environmental signals such as daylength and seasonal or day growth temperature, while the age and autonomous patways respond to endogenous signals, and the gibberellin pathway responds to both environmental and endogenous clues. All these pathways converge at the level of a few genes, named floral transition integrators.Within this group of floral transition integrators, several members of the MADS-box family have major roles: the expression of SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (SOC1) is activated by the photoperiod, age and gibberellin pathways to promote floral transition (Borner et al., 2000; Lee et al., 2000; Samach et al., 2000; Lee and Lee,© The Author 2014. Published by Oxford University Press on behalf of the Society for Experimental Biology.This is an Open Access article distributed under the terms of the Creative Commons Attribution License (/licenses/by/3.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. at Institute of Botany, CAS on May 4, 2014 / Downloaded from1194|Balanzà et al.2010) which is, in part, mediated by the activation of the floral identity gene LEAFY (LFY) (Lee et al., 2008; Liu et al., 2008). Conversely, FLOWERING LOCUS C (FLC) and SHORT VEGETATIVE PHASE (SVP) act as floral transition repres-sors (Hartmann et al., 2000; Michaels and Amasino, 1999; Sheldon et al., 1999). High levels of FLC expression compete the inductive floral signals at the SAM, and thus, flowering is promoted when the vernalization and autonomous path-ways repress FLC expression (Michaels and Amasino, 1999; Lee et al., 2000; Sheldon et al., 1999, 2000; Hepworth et al., 2002; Michaels et al., 2004; Kim et al., 2009). Likewise, the expression of the flowering repressor SVP is controlled by the autonomous, thermosensory, and gibberellin pathways (Lee et al., 2007; Li et al., 2008). FLC and SVP are able to form heterodimers that directly bind to the SOC1 promoter to down-regulate SOC1 expression, as well as to other floral transition integrators such as FLOWERING LOCUS T (FT) (Lee et al., 2007; Fujiwara et al., 2008; Li et al., 2008).The MADS-box transcription factor FRUITFULL (FUL), a closely related gene to the flower meristem identity genes APETALA1 (AP1) and CAULIFLOWER, has been associ-ated with several developmental processes. In addition to its well-known function during fruit development, FUL roles in floral meristem identity specification, shoot maturation, and the control of floral transition have also been described (Hempel et al., 1997; Gu et al., 1998; Ferrándiz et al., 2000a, b; Melzer et al., 2008; Shikata et al., 2009; Wang et al., 2009). FUL is partially redundant with SOC1 in flowering pro-motion. Although the ful mutants are only slightly late flow-ering under long-day growth conditions (Ferrándiz et al., 2000a), the double ful soc1 mutants show a strong delay in floral transition (Melzer et al., 2008). As SOC1, FUL is one of the earliest responsive genes to photoinductive signals (Hempel et al., 1997; Schmid et al., 2003) being a target of the FT–FD dimer (Schmid et al., 2003; Teper-Bamnolker and Samach, 2005; Torti et al., 2012). FUL also responds to signals derived from the age pathway, being one of the most responsive genes to the SQUAMOSA PROMOTER BINDING LIKE (SPL) proteins (Shikata et al., 2009; Wang et al., 2009; Yamaguchi et al., 2009). A recent study also places FUL in the promotion of flowering in response to ambient temperature through the action of miR156/SPL3 and FT (Kim et al., 2012).In spite of mounting evidence linking FUL to the main flowering pathways, the importance of FUL in controlling these processes, as well as its position, downstream effectors, and mode of action in these pathways are still unclear. In this study, genetic analyses have been used to understand better the regulatory hierarchies involving FUL and other floral integrators of the MADS-box family such as SOC1, SVP, and FLC in the control of floral transition in Arabidopsis. Our results show that FUL is able to act both upstream and co-operatively with SOC1, forming a heterodimer and bind-ing directly to the LFY promoter. In addition, it is shown that the promotive effect of FUL on floral transition depends of the presence of a functional allele of SVP and that FUL is able to counteract the repressive effect of FLC on flowering both affecting FLC expression levels and probably competing with FLC for common targets. Taking all these data together,a dynamic model is proposed for the role of FUL during flo-ral transition, where the progressive formation of different heterodimers of FUL and other MADS transcription fac-tors may act as a molecular switch between the vegetative and reproductive states.Materials and methodsPlant material and growth conditionsArabidopsis thaliana plants were grown in cabinets at 21 °C underLD (16 h light) or SD (8 h light) conditions, illuminated by cool-white fluorescent lamps (150 µE m–2 s–1), in a 1:1:1 by vol. mixtureof sphagnum:perlite:vermiculite. To promote germination, seedswere stratified on soil at 4 °C for 3 d in the dark. The Arabidopsisplants used in this work were in the Col-0 background, except ful-1and 35S::SOC1, that were in L er. Mutant alleles and transgenic lineshave been previously described: soc1-2 (Lee et al., 2000), ful-1 (Guet al., 1998), ful-2 (Ferrándiz et al., 2000a), svp-32 (Lee et al., 2007),FRI FLC (Lee and Amasino, 1995), 35S::SOC1, (Lee et al., 2000),35S::FUL (Ferrándiz et al., 2000b), 35S::SVP (Masiero et al., 2004),35S::FLC (Michaels and Amasino, 1999), LFY:GUS (Blázquezet al., 1997) and FLC:GUS (Sheldon et al., 2002).35S::FUL::GFP was generated by cloning the FUL CDS into the pEarley103 vector (Earley et al., 2006). Agrobacterium strain C58pM090 was used to transform Arabidopsis using the floral dip pro-tocol (Clough and Bent, 1998), and transgenic lines carrying a sin-gle transgene insertion and with similar phenotypes to the reference35S::FUL line were selected.Flowering time measurementsFlowering time was scored as number of leaves at bolting. The num-ber of rosette and cauline leaves was counted when the bolting shoothad produced the first open flower. At least 15 genetically identi-cal plants were used to score flowering time of each genotype. The Student’s t-test was used to test the significance of flowering time differences.Chromatin immunoprecipitation (ChIP)35S::FUL and 35S::FUL::GFP seeds were grown for 15 d in soiland inflorescences were collected for analysis. The ChIP experimentswere performed as previously described by Sorefan et al. (2009) withminor modifications using an anti-GFP antibody (Abcam, Ab290).Q-PCR was performed using the SYBR®Green PCR Master Mix (Applied Biosystems) in a ABIPRISM 7700 sequence detection system (Applied Biosystems). The values correspond to the ratios between the pull-down DNA with the GFP antibody from 35S::FULand 35S::FUL:GFP lines and between a 10% fraction of the input genomic DNA from both samples, all of them initially normalizedby ACT7 or UBQ10 genomic region. The primers used for this studyare described in Supplementary Table S1 at JXB online.Quantitative RT-PCR (qRT-PCR)Total RNA was extracted from whole plants with the RNeasy PlantMini kit (Qiagen). 2 µg of total RNA were used for cDNA synthesis performed with the First-Strand cDNA Synthesis kit (Invitrogen)and the qPCR master mix was prepared using the iQTM SYBR Green Supermix (Bio-Rad). Results were normalized to the expres-sion of the TIP41-like reference gene. The PCR reactions were runand analysed using the ABI PRISM 7700 Sequence detection system (Applied Biosystems). Three technical and two biological replicateswere performed for each sample. See Supplementary Table S1 atJXB online for the primer sequences.at Institute of Botany, CAS on May 4, 2014/Downloaded fromFUL modulates SVP and SOC1 activities |1195β-Glucuronidase (GUS) staining and activity measurementsFor GUS histochemical detection, samples were treated for 15 min in 90% ice-cold acetone and then washed for 5 min with washing buffer (25 mM sodium phosphate, 5 mM ferrocyanide, 5 mM ferri-cyanide, and 1% Triton X-100) and incubated from 4–16 h at 37 °C with staining buffer (washing buffer+1 mM X-Gluc). Following staining, plant material was fixed, cleared in chloral hydrate, and mounted to be viewed under bright-field microscopy.For quantitative measurements, the protocol described in Blazquez et al. (1997) was followed. Briefly, apices were incubated at 37 °C for 16 h in 1 mM MUG assay solution (1 mM 4-methyl umbelliferyl glucuronide, 50 mM sodium phosphate buffer pH 7, 10 mM EDTA, 0.1% SDS, 0.1% Triton X-100), in individual wells of a microtitre plate. After the reaction had been stopped by the addition of 0.3 M Na2CO3, fluorescence at 430 nm was measured on a luminescence spectrophotometer equipped with an ELISA plate reader (Perkin Elmer, model LS50B).Bimolecular Fluorescence Complementation (BiFC)Open reading frames of full-length FUL, SOC1, and SVP CDS were cloned into vectors pYFPN43 and pYFPC43 (http://www. ibmcp.upv.es/FerrandoLabVectors.php), and BiFC was performed as previously described by Belda-Palazon et al. (2012).Confocal microscopyConfocal microscopy was performed using a Leica TCS SL (Leica Microsystems GmbH, Heidelberg, Germany) equipped with an Argon krypton laser (Leica).Accession numbersSequence data from this article can be found in the Arabidopsis Genome Initiative or GenBank/EMBL databases under the follow-ing accession numbers: FUL (AT5G60910), SOC1 (AT2G45660), SVP (AT2G22540), FLC (AT5G10140), FRI (AT4G00650), LFY (AT5G61850), UBQ10 (AT4G05320), act7 (AT5G09810), and tip41-like (AT4G34270).ResultsGenetic interactions of FUL and SOC1The timing of both reproductive and meristem phase tran-sitions were compared by the quantification of rosette and cauline leaves of wild-type, ful, and 35S::FUL plants. As previously reported, it was observed that the loss of FUL function caused a small delay in flowering time both in long-day (LD) and short-day (SD) conditions, while the over-expression of FUL caused a strong early flowering phenotype (Table 1) (Ferrándiz et al., 2000a; Melzer et al., 2008). The late flowering phenotype of ful mutants mainly affected the onset of the meristem identity transition, since the number of rosette leaves did not significantly differ from the wild type, while the number of cauline leaves was increased in both LD and SD conditions (Table 1). In addition, when grown in SD, the axillary meristems of cauline leaves of single ful-2 mutants formed aerial rosettes (see Supplementary Fig. S1 at JXB online), and flowers were subtended by bracts (see Supplementary Fig. S1 at JXB online).It has been described that FUL and SOC1 have similar roles and probably promote flowering redundantly (Melzer et al., 2008). However, it is still unclear how precisely these two fac-tors interact genetically and how each of them contributes to the reproductive or the meristem identity transitions. To understand better the genetic relationship of FUL and SOC1, the effect on flowering time of different combinations of FUL and SOC1 loss- and gain-of-function alleles was compared. In LD conditions, the ful-2 soc1-2 double mutant showed a synergistic late-flowering phenotype, in agreement with pre-viously reported data (Melzer et al., 2008), producing more rosette leaves than the soc1-2 single mutant and more cauline leaves than both ful-2 and soc1-2 single mutants (Table 1). Additional phenotypes were observed such as the production of small leaves subtending flowers, the development of aerial rosettes at the cauline leaf axils, and frequent SAM rever-sion (see Supplementary Fig. S1B at JXB online), similar to what was observed in ful-2 single mutants grown in SD and in other studies (Torti et al., 2012).The soc1-2 mutant grown in SD showed a dramatic increase in rosette leaf number, and also a delay in meris-tem identity transition, although not as important as the delay produced by ful-2 (Table 1). The ful-2 soc1-2 double mutants grown in SD produced a similar number of rosetteTable 1. Genetic interaction of FUL and SOC1: effect on floweringLong day Short dayRosette leaves Cauline leaves Rosette leaves Cauline leavesColumbia-010.2 ± 1.0 3.2 ± 0.455.1 ± 3.49.3 ± 0.7ful-210.7 ± 0.8 4.4 ± 0.5a59.9 ± 3.8a23.7 ± 3.2asoc1-219.3 ± 0.9a 4.2 ± 0.5a75.0 ± 4.2a15.2 ± 0.5aful-2 soc1-224.5 ± 0.8a,b,c9.7 ± 1.9a.b,c75.1 ± 3.5a,b,28.1 ± 1.7a,b,c35S::FUL 3.5 ± 0.5a 1.7 ± 0.7a10.6 ± 0.9a 3.6 ± 0.7a35S::FUL soc1-29.0 ± 1.1d 2.2 ± 0.7d44.6 ± 12.8d7.2 ± 4.5d Landsberg er7.3 ± 0.5 1.8 ± 0.4nd ndful-18.4 ± 0.5e 2.5 ± 0.5e nd nd35S::SOC1 4.0 ± 0.0e0.4 ± 0.5e nd nd35S::SOC1 ful-1 4.0 ± 0.0f0.7 ± 0.5f,g nd nd35S::FUL 35S::SOC1 2.0 ± 0.0g0.2 ± 0.4g nd nd Flowering time is expressed as the mean of rosette and cauline leaves produced in long- and short-day conditions. Errors are represented as the standard deviation. Superscript letters indicate a significant difference (P <0.05) from (a) Col, (b) ful-2, (c) soc1-2, (d) 35S::FUL, (e) L er, (f) ful-1, and (g) 35S::SOC1 controls, respectively, according to Student’s t-test; nd=not determined. at Institute of Botany, CAS on May 4, 2014 / Downloaded from1196|Balanzà et al.leaves than the soc1-2 mutant, indicating that, in the absence of photoperiodic stimulus, the promoting role of FUL on the reproductive transition could depend on the presence of SOC1. On the other hand, the number of cauline leaves produced by ful-2 soc1-2 was only moderately higher than in ful-2 single mutants, suggesting that FUL would have a predominant effect in the control of meristem identity tran-sition (Table 1).35S::FUL soc1-2 plants flowered earlier than the wild type, but significantly later than 35S::FUL lines (Table 1) sup-porting the idea that the flowering-promoting role of FUL was partially dependent on the presence of an active allele of SOC1. In contrast, 35S::SOC1 ful-1 plants were iden-tical to 35S::SOC1 plants in rosette leaf number, while the absence of FUL only slightly increased the number of caul-ine leaves produced in the 35S::SOC1 background (Table 1). Finally, lines that over-expressed both genes simultaneously flowered extremely early, producing only two rosette leaves before the SAM directly differentiated into one or two flow-ers, although occasionally one cauline leaf with an axillary flower was formed (Table 1;Fig. 1A, B). Moreover, the axil-lary meristems from rosette leaves were also converted into flowers (Fig. 1A). This strong synergistic effect, together with the partial dependence of FUL on the presence of SOC1 to promote flowering, was compatible with FUL acting in part as an upstream regulator of SOC1, together with a subse-quent co-operative action of both proteins in the regulation of putative common targets, although it did not exclude other possible scenarios.SOC1 and LFY are FUL direct targetsIt has been described that FUL and SOC1 are able to inter-act in yeast two-hybrid experiments as homo- and heter-odimers (de Folter et al., 2005; Immink et al., 2012). To confirm this interaction in planta, a Bimolecular Fluorescence Complementation (BiFC) experiment was performed through transient expression on Nicotiana benthamiana leaves,observing FUL-SOC1 dimerization in the nuclei of the cells (Fig. 1C).The floral identity gene LFY has been identified as a bona fide SOC1 direct target (Lee et al., 2008). In addition, FUL has been also suggested to up-regulate LFY (Ferrándiz et al., 2000a). To confirm this suggestion, the expression of a LFY::GUS reporter line was analysed in the ful-2 and 35S::FUL backgrounds, and it was observed that the level of LFY expression was dependent on FUL, being lower in the ful-2 mutant and higher in the 35S::FUL line than in WT plants (Fig. 2A–C). These relative levels of expression were also confirmed by quantitative RT-PCR of LFY expression in apices at 7, 10, and 12 d after germination (Fig. 2D). In addition, GUS activity was also quantitatively determined in individual dissected apices, using the substrate 4-methyl umbelliferyl glucuronide (MUG), which is converted by GUS into the fluorescent product 4-MU. A time-course per-apex quantification was performed on the three genetic back-grounds, observing that LFY::GUS activity was consistently higher in 35S::FUL plants and lower in ful-2 plants than in the WT (Fig 2E). Chromatin immunoprecipitations (ChIP) experiments using a 35S::FUL::GFP line (see Supplementary Fig. S2 at JXB online) revealed that FUL was able to bind a region 2.2 kb upstream to the ATG codon of the LFY gene (Fig. 2F), overlapping with a previously identified region also bound by SOC1 (Lee et al., 2008).Moreover, FUL–GFP was also found to bind the SOC1 promoter, around 800 bp upstream of the ATG codon (Fig. 2G). Again, this region bound by FUL overlaps witha region bound by SOC1 itself, which confirms in planta theY1H experiment reported previously, which shows a FUL–SOC1 heterodimer binding to this fragment of the SOC1 promoter (Immink et al., 2012). Taken together, these results strongly support the hypothesis of SOC1 and FUL bindingas heterodimers to the promoters of their target genes and could explain the genetic interactions observed.Genetic interactions of FUL and SVPSVP has been shown to repress SOC1 directly, in part by binding to the SOC1 promoter as a heterodimer with FLC,a potent repressor of flowering involved in the vernalizationFig. 1. Interaction of FUL with SOC1. (A, B) Phenotypes of 35S::FUL35S::SOC1 double over-expression lines. Only two rosette leaves areproduced (arrows in A) and occasionally one cauline leaf (arrowhead inB). All axillary meristems are determinate, directly producing flowers.Asterisks mark the cotyledons in (A). (C) Bimolecular Fluorescence Complementation in tobacco epidermal leaf cells between transiently expressed FUL and SOC1 fusions to the C- and N-terminal fragments ofYFP, respectively. The left panel shows reconstituted YFP fluorescence (green) and the right panel is an overlay with a bright field image of thesame sector where chlorophyll is shown in red. Negative controls for BiFC experiments are shown in Supplementary Fig. S3 at JXB online. Scale bars: 500 mm (A, B), 40 µm (C). at Institute of Botany, CAS on May 4, 2014/Downloaded fromFUL modulates SVP and SOC1 activities | 1197and autonomous pathways (Michaels and Amasino, 1999; Sheldon et al., 2002; Helliwell et al., 2006). Our results indicated that FUL could also act as an upstream regu-lator of SOC1, binding directly the SOC1 promoter. To explore whether FUL could interact with SVP to regulate SOC1, the effect on flowering time of different combina-tions of FUL and SVP loss- and gain-of-function alleleswas characterized.Fig. 2. FUL regulates key genes in the floral transition process binding directly to SOC1 and LFY promoters. (A–C) Histochemical detection of LFY::GUS activity in the apices of 6-d-old wild type (A), ful-2 (B) or 35S::FUL (C) plants. Scale bars, 250 µm. (D) Relative expression of LFY analysed by qRT -PCR in WT, ful-2, and 35S::FUL plants at 7, 10, and 12 d after germination. The error bars depict the s.e. based on two biological replicates. Asterisks (*) indicate a significant difference (P <0.05) from the WT control according to Student’s t -test. (E) Quantification of LFY:GUS activity in WT, ful-2, and 35S::FULbackgrounds. Plants were grown on plates under long days (LD). At each time point, GUS activity was measured in at least 12 individual apices, and the means ±s.e are given. (F) (Top) Schematic diagram of the LFY upstream promoter region. First exon is represented by a black box, while the upstream genomic region is represented by a black line. The red stars indicate the sites containing either single mismatch or perfect match with the consensus binding sequence (CArG box) of MADS-domain proteins. Amplicons spanning these sites used in the ChIP analyses are represented by grey lines and marked by roman numbers. (Bottom) ChIP enrichment tests showing the binding of FUL-GFP to the LFY-I region. Bars represent the ratio of amplified DNA (35S::FUL:GFP/35S::FUL) in the starting genomic DNA (input) or in the immunoprecipitated DNA with the GFP antibody (Ab). (G) (Top) Schematic diagram of the SOC1 genomic region, including upstream promoter, exons 1 and 2 and the first intron. Exons are represented by black boxes, upstream genomic region and intron by a black line. The red stars mark CArG boxes. Amplicons spanning these sites used in the ChIP analyses are represented by grey lines and marked by roman numbers. (Bottom) ChIP enrichment tests showing the binding of FUL-GFP to the SOC1-III region. Bars represent the ratio of amplified DNA (35S::FUL:GFP/35S::FUL) in the starting genomic DNA (input) or in the immunoprecipitated DNA with the GFP antibody (Ab).at Institute of Botany, CAS on May 4, 2014/Downloaded from1198 | Balanzà et al .The svp-32 mutant showed a clear early-flowering pheno-type both in LD and SD conditions, reducing the number of rosette leaves produced when compared with the WT control, as previously described by Lee et al. (2007) (Table 2). ful-2 svp-32 flowered with a similar number of leaves as the svp-32 single mutant (Table 2) (Torti et al., 2012), suggesting that SVP represses additional targets that can promote flower-ing in the absence of FUL , as has already been proposed by Torti et al. (2012). If this was true, we could expect plants over-expressing FUL in a svp background to flower earlier or at least like 35S::FUL plants. However, 35S::FUL svp-32 plants also flowered similarly to svp-32, both in LD and SD, (Table 2) suggesting an alternative scenario where FUL over-expression was not able to promote flower transition in the absence of an active SVP protein. Thus, the epistatic effect of svp mutation on both FUL loss- or gain-of-function may sug-gest that FUL required SVP to regulate its targets, and this could be mediated by the physical interaction of both factors.Interaction of FUL and SVP proteins has already been reported in yeast-two-hybrid experiments (de Folter et al., 2005; Immink et al., 2012). To test if this heterodimer also occurred in planta , a BiFC experiment was performed that confirmed such interaction (Fig 3A ). If FUL required inter-action with SVP to promote floral transition, it could be expected that simultaneous over-expression of FUL and SVP would result in early flowering, overcoming the late-flower-ing phenotype caused by SVP over-expression. A 35S::SVP 35S::FUL line was then generated and flowering time quan-tified in this double transgenic line. As described above, 35S::FUL flowered early, while 35S::SVP flowered very late, as expected for a potent repressor of flowering transition (Table 2; Fig. 3B ). The line harbouring both the 35S::FUL and the 35S::SVP transgenes flowered early, similarly to 35S::FUL or 35S::FUL svp plants (Fig. 3B ; Table 2). This phenotype indicated that SVP was not able to repress floral transition when both high levels of SVP and FUL were pre-sent, suggesting that the FUL–SVP dimer could suppress the repressor effect of SVP on flowering or even act as a flowering promoting factor.Genetic interactions of FUL and FLCBecause the repressor effect of SVP in flowering transition is partially mediated by the formation of a heterodimer with FLC (Lee et al., 2007; Fujiwara et al., 2008; Li et al., 2008), the genetic relationship of FUL and FLC was studied.Much of the natural variation in flowering time in Arabidopsis depends on the allelic variation of FLC andTable 2. Genetic interaction of FUL and SVP: effect on floweringLong day Short dayRosette leavesCauline leavesRosette leavesCauline leavesColumbia-012.4 ± 1.7 2.5 ± 0.464.4 ± 6.08.6 ± 0.8ful-212.9 ± 0.9 3.8 ± 0.6a 70.2 ± 7.0a 20.8 ± 3.8a svp-325.6 ± 0.5a 2.8 ± 0.416.4 ± 2.1 4.6 ± 1.0ful-2 svp-32 5.3 ± 0.5b 3.3 ± 0.516.1 ± 2.57.1 ± 1.635S::FUL4.0 ± 0.0a 1.4 ± 0.5a 8.3 ± 1.8a 3.5 ± 0.8a 35S::FUL svp-325.8 ± 0.4 2.5 ± 0.514.9 ± 2.1c,d 3.4 ± 1.2c 35S::SVP27.5 ± 1.7a 7.3 ± 1.0a nd nd 35S::FUL 35S::SVP5.8 ± 1.2e2.7 ± 0.8d,endndFlowering time is expressed as the mean of rosette and cauline leaves produced in long- and short-day conditions. Errors are represented as the standard deviation. Superscript letters indicate a significant difference (P <0.05) from (a) Col, (b) ful-2, (c) svp-32, (d) 35S::FUL, and (e) 35S::SVP controls, respectively, according to Student’s t-test; nd=not determined.Fig. 3. Interaction of FUL with SVP . (A) BiFC experiments in tobacco leaf cells between transiently expressed FUL and SOC1 fusions to the C- and N-terminal fragments of YFP , respectively. The left panel shows YFP reconstituted fluorescence (green) and the right panel is an overlay with a bright field image of the same sector where chlorophyll is shown in red. Negative controls for BiFC experiments are shown in Supplementary Fig. S3 at JXB online. Scale bars: 40 µm. (B) Phenotypes of the 35S::FUL, 35S::SVP , and 35S::FUL 35S::SVP double over-expression lines. FULover-expression reverts the late flowering phenotype of 35S::SVP , although inflorescence development is partially restored respect to the 35S::FUL plants.at Institute of Botany, CAS on May 4, 2014/Downloaded from。
G E N E R A L I N F O R M A T I O NGio provides complete control of conventional and moving lights, LEDs and media servers. Supports multiple users with partitioned parameter control and full backup, multiple playback faders and cue lists in a tracking, move fade environment, with unique force-feedback encoders, two integral articulating multi-touch displays and backlit keys.FEATURES• 4,096 or 24,576 outputs• 32,768 control channels• Up to 99 discrete users• Partitioned control• Master playback pair with motorized 100mm faders• T en 100mm motorized faders x 100 pages for configurablecue lists, submasters, grand masters, IFCB Palette/Presets listsor individual instances• T wo 12.1 inch multi-touch LCD touchscreens for display,direct selection and context-sensitive control• Four discrete palette types (IFCB)• Presets function as “all palette”• E ffects provide dynamic relational and absolute progressivebehavior• C entral information area (CIA) accesses electronic alphakeyboard, Hue + Saturation color picker, gel picker, browserand other controls• F our force-feedback encoders for non-intensityparameter control• C onfigurable high-density channel display, with format andflexi-channel modes• User-designed, interactive magic sheets• U p to six abstract color spaces, tinting, spectrum and fadepath tools.• E TCNet2™ and Net3™ (powered by ACN), ArtNet and Avab®UDP output protocols• S how import from Obsession, Express™, Expression®,Emphasis®, Congo®, Cobalt®,Grand MA1, Grand MA2, Safariand Strand 500/300 Series• T wo individually configurable Ethernet ports• M ultiple MIDI and/or SMPTE TimeCode Inputs, MIDI In andOut, Analog/Serial Inputs, OSC transmit/receive• V irtual Media Server function for pixel-mapped effects,images and animations• S upport for multiple languages, including English, German,Spanish, French, Italian, Japanese, Korean, Russian andChinese (Simplified and Traditional)O R D E R I N G I N F O R M A T I O NGioMODEL DESCRIPTIONGio – 4K Gio console, 4,096 outputs (minimum)Gio – 24K Gio console, 24,576 outputs (maximum) Eos RPU3 – 4K Eos Remote Processor Unit, 4,096 outputs Eos RPU3 – 24K Eos Remote Processor Unit, 24,576 outputs Gio 20K Up After-sale 20K upgrade (display port) ETCnomad 512Client for PC/MacO utput protocols are distributed using ETCNet2 DMX Nodes or Net3 DMX/RDM Gateways. I/O Gateways and Show Control Gateways provide switch closure functionality, MIDI and SMPTE TimeCode.Note: Gio with three display port outputs on the rear facepanel can be upgraded to 24,576 outputs. Gio with three DVI ports can be upgraded to 12,288 outputs. Use Gio 10K up to increase the output capacity of those units to maximum capacity. For projects exceeding 24K of output, please contact ETC.Gio AccessoriesMODEL DESCRIPTIONEos FW 20Eos Standard Fader Wing 20Eos FW 40Eos Standard Fader Wing 40Eos MFW10Eos Motorized Fader Wing 10Eos MFW20Eos Motorized Fader Wing 20Net3 RVI3Remote Video InterfaceETCpad ETC Portable Access DeviceGIO – FC Gio FlightcaseEos Family Offline Editor software for Mac and PC platforms is called ETCnomad and is available for download from Gio requires Windows 7 compatible external monitors, 1280x1024 minimum resolution, standard, touch or multi-touchSHIPS WITH:• Dust cover• Two Littlites• Mouse and mousepad• Backlit external alphanumeric keyboard•Three active display port to DVI adaptersS P E C I F I C A T I O N SSYSTEM CAPACITY• 4,096 or 24,576 outputs• 32,768 Control Channels (devices)• 10,000 Cues• 999 Cue Lists• 200 Active Playbacks• 999 Submasters• 100 Fader Pages• 4 x 1,000 Palettes (Intensity, Focus, Color, Beam)• 1,000 Presets (all palette)• 1,000 Groups• 1,000 Effects (relative, absolute or step)• 99,999 Macros• 1,000 Snapshots• 1,000 Curves• 1,000 Color Paths• S upports three external display port monitors at 1280 x 1024(minimum resolution required with optional touch or multi-touch control)• Solid-state hard drive• Seven USB ports for flashdrives, pointing devices, keyboards DISPLAY FUNCTIONS• A ll show data may be viewed on a single external monitor ormay be posted to the integral touchscreens. External viewsmay be posted separately or expanded across a maximumof three monitors. Three user-configurable workspaces perdisplay, with split-screen/sizing controls.• The Central Information Area accesses:-Browser-File Management-System Defaults-Show Defaults-Desk Defaults-Partition Definitions-Network Configuration-Show Data Utilities-Print to PDF-Record Target Lists-Help-Electronic alpha keyboard-Command Line-Selected Cue-Error messages-Context Sensitive Control-Parameter Categories and individual parameters-Filters• Channel Displays-Live channel or table view-B lind cue, palette, preset and group views, in list, channel,table and spreadsheet formats-U ser-configurable to show required parameters and/orparameter categories (IFCB)-Flexi-channel to determine which channels to display-Zoom allows user to define how many channels are viewed-Color-coded intensity levels indicate direction of move-Option to display referenced by number or label-C olor-coded non-intensity levels indicate change fromprevious state-G raphic differentiation of moving lights, single parameterdevices and unpatched channels S P E C I F I C A T I O N S• Magic Sheets-User-defined interactive display layouts-Objects and images may be imported• Patch Views-Patch by channel-Patch by address-Patch by Device List (RDM)-A ssign proportional patch value, curve, preheat valuefor intensity-Swap/Invert pan and tilt-User configurable shutter order-Custom fixture editor• Playback Status Display-Accesses status of 30 fader pages-E xpanded cue list for selected cue. Optional dynamiccountdown of active cues-Order/hide content per instance• Cue List Index• Effect Editor• Group Editor• Park Display• Dimmer Monitoring• Submaster listPLAYBACK CONTROLS• M aster Playback crossfade pair with two 100mm (3.94in.)motorized potentiometers, user-configurable button/sliderbehavior• 100 pages of ten 100mm (3.94in.) motorized faders, eachconfigurable as:-IFCB Palette/Presets Lists or single instances-S ingle playback, with user-configurable button/sliderbehavior-Grand Master with Blackout-A dditive or Inhibitive Submaster, with user-configurablebutton/slider behavior-Filtered Manual Timing Master• Rate controller• Playback fader controls include:-Load to assign cue lists-Timing Disable-Off/On-Release-Filters-Freeze-Assert-Manual Override-Rate-Go To Cue 0-Spread-Background enabled/disabled-10 Priority States-10 Background Priority States-Parameter and channel filtersMACROS• May be set to play background or foreground• Startup and Shutdown Macros• Disconnect MacrosS P E C I F I C A T I O N SMANUAL CONTROL• C hannel selection from keypad and/or direct selects• Lists constructed with +, -, thru• I ntensity set with level wheel, keypad, level button, fulland out• Select Last recalls last sequential channel selection set• Select Manual selects all channels with manual values• Select Active selects all channels with intensity above zero• Ordered groups• Offset; including even, odd, random and reverse• Fan• Sneak• User-definable home• H ome by parameter, parameter category or all non-intensityparameters• Capture• Park at level• Scaled park for temporary percentage adjustment• Recall-from and copy-to commands• A bout provides detailed view of selected channels orrecord targets• Undo• Highlight and Lowlight, with optional user-definable Rem Dim • L amp controls to strike and douse arc sources,calibrate devicesPROGRAMMING FEATURES• Channel Functions-N on-intensity parameters set via numeric entry orpageable encoders-Encoders support software-controlled tactile response-Local display of color and gobo images-C olor matching to gel selector-Color Path, color tinting and color spectrum tools-Apply discrete time and delay per channel parameter• Palette and Preset Functions-Record and Update-Toggle display to absolute data-U p to 99 decimal values may be inserted between any twowhole numbers• Effects-Create live or blind-Pattern-based relative dynamic effects-Absolute effects-Step effects-Channel level overrides-Cue level overrides-Entry mode determines how parameters enter effects-Exit mode determines how parameters depart effects• Cue Recording-Cue List HTP/LTP Intensity-Cue List Priority and Background Priority-Cue List Assert-Fader as progress controller, manual or intensity master-Record manual values or channels in use-Auto playback of recorded cues-Referenced or auto-mark instructions-Block at cue or parameter level S P E C I F I C A T I O N S-Assert at cue or parameter level-All-fade flag-Follow or hang times-Out of sequence link-Loop functions-Cue level parameter category timing-20-part multi-part cues with default part assignment-Cue-level rate override-Mark flags for Auto or Referenced Marks-U p to 99 decimal cues between each twowhole-numbered cues-Execute List·Triggers snapshot·Triggers macros·Triggers go of other cues·Syncs go to multiple cue lists·Show-control triggers·Analog triggers-Update and Update Trace functions-Undo record and delete• Submaster Recording and Playback-999 additive or inhibitive submasters-Bump button timing for fade up/dwell/fade out-Assert/Channel select button-Exclusive or Shielded Mode-Background enable/disable-Restore to background or minimum value-LTP/HTP intensity-Fader as progress controller or intensity master-Bump button to mark NPs-Priority and Background Priority status-M otorized faders match level across all devices andwhen paging-Submaster mapping on the fly• Curves-Assignable in patch to modify dimmer output ramp-A ssignable at cue or cue part level to modify intensitycrossfade profile or non-intensity parameter ramping INTERFACES• Ethernet (two ports) 802.3af compliant PSE• E TCNet2, Net3 (powered by ACN), ArtNet and Avab UDPoutput protocols• Four DMX/RDM ports• Contact-closure triggers via D-Sub connector• T hree video connectors support display port external displays(1280x1024) with optional single-touch or multi-touchscreen control• USB multipurpose bus (seven ports)• OSC Transmit/Receive• MIDI In/Out (MIDI TimeCode, MIDI Show Control)• SMPTE TimeCode through Gateway• C ontact closure (12 analog inputs, 12 SPDT contact outputs,RS-232) through GatewayELECTRICAL• AC input (100 - 240V at 50/60 Hz)• P ower consumption (less external monitors) approximatelytwo amps at 120V and one amp at 230/240VP H Y S I C A L Gio Dimensions *MODEL HEIGHT WIDTH DEPTH inches mm inches mm inches mm Gio11.6129530.5774.723.2589.28Gio in shipping container 34.5876.327.3692.210.1257.2Gio in roadcase34863.636.2919.59.3234.2Gio Weight*MODELWEIGHTlbs kgs Gio console4520.5Gio in shipping container 6027.2Gio in roadcase8036.3*Weight and dimensions typical7.87mmCorporate Headquarters 3031 Pleasant View Rd, PO Box 620979, Middleton WI 53562 0979 USA +1 608 831 4116London, UK Unit 26-28, Victoria Industrial Estate, Victoria Road, London W3 6UU, UK +44 (0) 20 8896 1000Rome, IT Via Pieve Torina, 48, 00156 Rome, Italy +39 (06) 32 111 683Holzkirchen, DE Ohmstrasse 3, 83607 Holzkirchen, Germany +49 (80 24) 47 00-0Hong Kong Room 1801, 18/F , Tower 1 Phase 1, Enterprise Square, 9 Sheung Yuet Road, Kowloon Bay, Kowloon, Hong Kong +852 2799 1220Web Copyright©2018 ETC. All Rights Reserved. All product information and specifications subject to change. 4240L1012 Rev K 02/18。
DPSP:Distributed Progressive SequentialPattern Mining on the CloudJen-Wei Huang1,Su-Chen Lin2,and Ming-Syan Chen21Yuan Ze University,Taiwanjwhuang@.tw2National Taiwan University,TaiwanAbstract.The progressive sequential pattern mining problem has beendiscussed in previous research works.With the increasing amount ofdata,single processors struggle to scale up.Traditional algorithms run-ning on a single machine may have scalability troubles.Therefore,miningprogressive sequential patterns intrinsically suffers from the scalabilityproblem.In view of this,we design a distributed mining algorithm to ad-dress the scalability problem of mining progressive sequential patterns.The proposed algorithm DPSP,standing for Distributed Progressive Se-quential Pattern mining algorithm,is implemented on top of Hadoopplatform,which realizes the cloud computing environment.We proposeMap/Reduce jobs in DPSP to delete obsolete itemsets,update currentcandidate sequential patterns and report up-to-date frequent sequentialpatterns within each POI.The experimental results show that DPSPpossesses great scalability and consequently increases the performanceand the practicability of mining algorithms.1IntroductionBased on the earlier work[7],the sequential pattern mining problem[1]can be categorized as three classes according to the management of correspond-ing databases.They are static sequential pattern mining,incremental sequential mining and progressive sequential pattern mining.It is noted that the progressive sequential pattern mining is known as a general model of the sequential pattern mining.The static and the incremental sequential pattern mining can be viewed as special cases of the progressive sequential pattern mining.The progressive se-quential pattern mining problem can be described as“Given an interesting time period called period of interest(POI)and a minimum support threshold,find the complete set of frequent subsequences whose occurrence frequencies are greater than or equal to the minimum support times the number of sequences having elements in the current POI in a progressive sequence database.”In fact,mining progressive sequential patterns intrinsically suffers from the scalability problem. In this work,we propose a distributed data mining algorithm to address the scalability problem of the progressive sequential pattern mining.The proposed algorithm DPSP,which stands for Distributed Progressive Sequential Pattern mining algorithm,is designed on top of Hadoop platform[6],which implements Google’s Map/Reduce paradigm[5].M.J.Zaki et al.(Eds.):PAKDD2010,Part II,LNAI6119,pp.27–34,2010.c Springer-Verlag Berlin Heidelberg201028J.-W.Huang,S.-C.Lin,and M.-S.ChenWe design two Map/Reduce jobs in DPSP.At each timestamp,the candi-date computing job computes candidate sequential patterns of all sequences and updates the summary of each sequence for the future computation.Then,us-ing all candidate sequential patterns as the input data,the support assembling job accumulates the occurrence frequencies of candidate sequential patterns in the current POI and reports frequent sequential patterns to users.Finally,all up-to-date frequent sequential patterns in the current POI are reported.DPSP not only outputs frequent sequential patterns in the current POI but also stores summaries of candidate sequential patterns at the current timestamp.As time goes by,DPSP reads back summaries of all sequences and combine them with newly arriving itemsets to form new candidate sequential patterns at the new timestamp.Obsolete candidate sequential patterns are deleted at the same time. DPSP is thus able to delete obsolete itemsets,update summaries of all sequences and report up-to-date frequent sequential patterns.It is noted that DPSP does not need to scan the whole database many times to gather occurrence frequencies of candidate sequential patterns.DPSP,instead,reads newly arriving data and the summary of each sequence once.In addition,DPSP utilizes cloud comput-ing techniques.It is easy to scale out using Hadoop platform to deal with huge amounts of data.The experimental results show that DPSP canfind progressive sequential patterns efficiently and DPSP possesses great scalability.The dis-tributed scheme not only improves the efficiency but also consequently increases the practicability.The rest of this work is organized as follows.We will derive some preliminaries in Section2.The proposed algorithm DPSP will be introduced in Section3.Some experiments to evaluate the performance will be shown in Section4.Finally,the conclusion is given in Section5.2Related WorksAfter thefirst work addressing the sequential pattern mining problem in[1], many research works are proposed to solve the static sequential pattern min-ing problem[2],and the incremental sequential pattern mining problem[10]. As for the progressive sequential pattern mining problem,new data arrive at the database and obsolete data are deleted at the same time.In this model, users can focus on the up-to-date database andfind frequent sequential patterns without being influenced by obsolete data.To deal with a progressive database efficiently,a progressive algorithm,Pisa,is proposed in[7].However,traditional algorithms running on a single processor struggle to scale up with huge amount of data.In view of this,many researchers work on distributed and parallel data mining algorithms[4][3][12][9][11][8].In recent days,many researchers and corporations work on developing the cloud computing technology,which utilizes clusters of machines to cope with huge amount of data.The platform allows developers to focus on designing distributed algorithms whereas routine issues like data allocation,job scheduling,load balancing,and failure recovery can be inherently handled by the cloud computing framework.Hadoop[6]is an openDPSP:Distributed Progressive Sequential Pattern Mining on the Cloud29˜́̃̈̇ˍ˼̇˸̀̆˸̇̆ʳ̂˹ʳ˴˿˿ʳ̆˸̄̈˸́˶˸̆ʳ˴̅̅˼̉˼́˺ʳ˴̇̇˻˸˶̈̅̅˸́̇ʳ̇˼̀˸̆̇˴̀̃˄ˁʳ̊̊˻˼˿˸ʻ̇˻˸̅˸˼̆́˸̊ʳ˷˴̇˴ʳ˴̇ʳ̇˼̀˸̆̇˴̀̃ʳ̇ʼ̎˅ˁʳ˖˴́˷˼˷˴̇˸˖̂̀̃̈̇˼́˺˝̂˵ˎˆˁʳ˦̈̃̃̂̅̇˔̆̆˸̀˵˿˼́˺˝̂˵ˎˇˁʳ̇ʳːʳ̇ʳʾʳ˄ˎˈˁʳ̂̈̇̃̈̇ʳ˹̅˸̄̈˸́̇ʳ̆˸̄̈˸́̇˼˴˿ʳ̃˴̇̇˸̅́̆ˎˉˁʳ̐˸˸́˷̊˻˼˿˸̂̈̇̃̈̇ˍʳˣˢ˜̊˼̇˻̇˻˸˼̅̆̈̃̃̂̅̇̆˘́˷Fig.1.Algorithm DPSP and system modelsource project aiming at building a cloud infrastructure running on large clusters, which implements Google’s Map/Reduce paradigm[5].By means of the map function,the application can be divided into several fractions.Each fraction is assigned to a single node in large clusters and executed by the node.After the execution,the reduce function merges these partial results to form thefinal output.As such,developers need only to design a series of Map/Reduce jobs to split data and merge results.3Distributed Progressive Sequential Pattern MiningWe utilize Hadoop platform to design a distributed algorithm for the progressive sequential pattern mining.The proposed algorithm is named as Distributed Pro-gressive Sequential Pattern mining algorithm,abbreviated as DPSP.In essence, DPSP consists of two Map/Reduce jobs,the candidate computing job and the support assembling job.As shown in the left of Figure1,for each timestamp,the candidate computing job reads input data,which arrives at timestamp t,of all sequences.Itemsets from different sequences are distributed to different nodes in the cloud computing environment.Each node in the cloud computes candidate sequential patterns of each sequence within the current POI.Meanwhile,the candidate computing job also updates the summary for each sequence.Obsolete data are deleted in the candidate computing job and the up-to-date candidate sequential patterns are output.Then,support assembling job reads all candidate sequential patterns as input data.Different candidate sequential patterns are dis-tributed to different nodes.Each node accumulates the occurrence frequencies of candidate sequential patterns and reports frequent sequential patterns whose supports are no less than the minimum support threshold in the current POI30J.-W.Huang,S.-C.Lin,and M.-S.Chento users.When time goes to the next timestamp,DPSP keeps executing these Map/Reduce jobs.As such,DPSP is able to report the most up-to-date frequent sequential patterns in each POI.The system model of DPSP is shown in right of Figure1.The upper part is the candidate computing job while the support assembling job is at the lower part. In the candidate computing job,input data at timestamp t and the candidate set summaries at timestamp t-1are split and transferred to several Mapper generates many pairs of<sequence number,input itemset>. Then,pairs with the same sequence number are sent to the same CCReducer. CCReducer computes candidate sequential patterns of the given sequence and outputs pairs of<candidate sequential patterns,null>.In addition,CCReducer updates the summary of each sequence and deletes obsolete data at the same time.Candidate set summaries at the current timestamp are output for the com-putation at the next timestamp as well.Next,each SAMapper in the support assembling job reads input data and accumulates local occurrence frequencies for each candidate sequential patterns.SAMapper generates pairs of<candidate sequential pattern,local supports of the candidate>as outputs.Then,the pairs containing the same candidate sequential pattern are sent to the same SARe-ducer.SAReducer aggregates supports of the same candidate sequential pattern and outputs those frequent patterns in the current POI.After the computation at the timestamp,t,DPSP moves to the next timestamp,t+1.3.1Candidate Computing JobThe objective of the candidate computing job is to compute all candidate se-quential patterns from all sequences within the current POI as shown in Figure2. In CCMapper,itemsets of all sequences arriving at the current timestamp and the candidate set summaries at the previous timestamp are used as input data. As shown in lines2to3of CCMapper,if CCMapper reads the input from candi-date set summaries,CCMapper generates<sequence number,candidate itemset with the corresponding timestamp>pairs.On the other hand,if CCMapper reads the input data from a sequence,CCMapper outputs<sequence number, arriving itemset>pairs as shown in lines4to5.These output pairs are dis-tributed to CCReducers as their inputs.Pairs with the same key are sent to the same CCReducer.By means of the summary at the previous timestamp and the arriving itemset at the current timestamp,each CCReducer is able to gen-erate candidate sequential patterns of each sequence in the current POI.In line 2of CCReducer,the multiple output variable is used to output candidate set summary at the current timestamp for the future computation.In lines6to15, CCReducer enumerates each value in the receiving pairs.If the value is a can-didate set summary at the previous timestamp,CCReducer puts the candidate into cand set.In lines9to10,if the timestamp is bigger than the start time of the current POI,which means this candidate will still be valid at the next timestamp,CCReducer outputs the candidate in the summary of the current timestamp for the computation at the next timestamp.In lines11to12,if the candidate contains more than1item,the candidate is put in the result set asDPSP:Distributed Progressive Sequential Pattern Mining on the Cloud31˖˖ˠ˴̃̃˸̅˼́̃̈̇ˍʳ˼̇˸̀̆˸̇̆ʳ̂˹ʳ˴˿˿ʳ̆˸̄̈˸́˶˸̆ʳ˴̅̅˼̉˼́˺ʳ˴̇̇˻˸˶̈̅̅˸́̇̇˼̀˸̆̇˴̀̃ˢ˥˶˴́˷˼˷˴̇˸̆˸̇̆̈̀̀˴̅˼˸̆˴̇̇˻˸̃̅˸̉˼̂̈̆̇˼̀˸̆̇˴̀̃̀˴̃ˍ˄ˁʳ̉˴̅ʳ˷˴̇˴ʳːʳ̅˸˴˷ʳ˼́̃̈̇ʳ˷˴̇˴ˎ˅ˁ˼˹ʻ˷˴̇˴˶̂́̇˴˼́̆̇˼̀˸̆̇˴̀̃ʼ̎˂˂˼̇˼̆˴˶˴́˷˼˷˴̇˸ˆˁ̂̈̇̃̈̇ʳˏ˷˴̇˴ˁ̆˸̄̈˸́˶˸ˡ̂ʿʳ˷˴̇˴ˁ˼̇˸̀̆˸̇ʾ˷˴̇˴ˁ̇˼̀˸ˑˎˇˁ˸˿̆˸˂˂˼̇˼̆˴́˴̅̅˼̉˼́˺˼̇˸̀̆˸̇ˈˁʳ̂̈̇̃̈̇ʳˏ˷˴̇˴ˁ̆˸̄̈˸́˶˸ˡ̂ʿʳ˷˴̇˴ˁ˼̇˸̀̆˸̇ˑˎ˄ˌˁʳʳʳʳ̅˸̆̈˿̇ˁ̃̈̇ʻ́˸̊˲˶˴́˷ʼˎ˅˃ˁʳʳʳʳ˼˼˹ʻ˶˴́˷˼˷˴̇˸ˁ̇˼̀˸ʳˑʳ̆̇˴̅̇˲̇˼̀˸ʼ˅˄ˁʳʳʳʳʳʳ̀̂ˁ̂̈̇̃̈̇ˏ˼́˲˾˸̌ʿʳ́˸̊˲˶˴́˷ʳʾʳ˶˴́˷˼˷˴̇˸ˁ̇˼̀˸ˑˎ˅˅ˁʳʳ̐˸˸́˷ʳ˹̂̅˅ˆˁʳʳ̀̂ˁ̂̈̇̃̈̇ˏ˼́˲˾˸̌ʿʳ˶̂̀˵˼́˴̇˼̂́ʳʾ˶̈̅̅˸́̇ʳ̇˼̀˸ˑˎ˅ˇˁ̐˸˸́˷ʳ˹̂̅˅ˈˁ˹˹̂̅ʻ˸˴˶˻ʳ˼̇˸̀̆˸̇ʳ˼́ʳ̅˸̆̈˿̇ʼ˅ˉˁʳʳ̂̂̈̇̃̈̇ˏ˼̇˸̀̆˸̇ʿʳ́̈˿˿ˑˎ̂̈̇̃̈̇ˍʳ̆˸̄̈˸́˶˸ʳ́̈̀˵˸̅ʳ˴́˷ʳ˶˴́˷˼˷˴̇˸ʳ̆˸̄̈˸́̇˼˴˿ʳ˼̇˸̀̆˸̇̆ʳ̊˼̇˻ʳ̇˼̀˸̆̇˴̀̃̆˦˔ˠ˴̃̃˸̅˼́̃̈̇ˍʳ˶˴́˷˼˷˴̇˸ʳ̆˸̈̄˸́̇˼˴˿ʳ̃˴̇̇˸̅́̆ʳ̂˹ʳ˴˿˿ʳ̆˸̄̈˸́˶˸̆˶̂́˹˼˺̈̅˸ˍ˄ˁʳʳ̉˴̅ʳ˿̂˶˴˿ˠ˴̃ˏ˼̇˸̀̆˸̇ʿʳ̆̈̃̃̂̅̇ˑˎʳʳʳʳ˂˂ʳ̈̆˸˷ʳ̇̂ʳ˿̂˶˴˿˿̌ʳ˴˺˺̅˸˺˴̇˸ʳ̆̈̃̃̂̅̇̆ˁ̀˴̃ˍ˅ˁʳʳ̉˴̅ʳ˷˴̇˴ʳːʳ̅˸˴˷ʳ˼́̃̈̇ʳ˷˴̇˴ˎˆˁʳʳ˼˼˹ʻ˷˴̇˴ˁ˼̇˸̀̆˸̇ʳ˼̆ʳ˼́ʳ˿̂˶˴˿ˠ˴̃ʼˇˁʳʳʳʳʳ˿̂˶˴˿ˠ˴̃ʻ˼̇˸̀̆˸̇ʼˁ̆̈̃̃̂̅̇ʾʾˎˈˁʳʳ˸˸˿̆˸ˉˁʳʳʳʳʳ˼́̆˸̅̇ʳˏ˼̇˸̀̆˸̇ʿʳ˄ˑʳ˼́̇̂ʳ˿̂˶˴˿ˠ˴̃ˎ˶˿̂̆˸ˍˊˁʳʳ˹˹̂̅ʻ˸˴˶˻ʳˏ˼̇˸̀̆˸̇ʿ̆̈̃̃̂̅̇ˑ̃˴˼̅ʳ˼́ʳ˿̂˶˴˿ˠ˴̃ʼˋˁʳʳʳʳ̂̂̈̇̃̈̇ʳˏ˼̇˸̀̆˸̇ʿʳ̆̈̃̃̂̅̇ˑˎ˦˔˥˸˷̈˶˸̅̅˸˷̈˶˸ʻ˼́˲˾˸̌ʿ˼́˲̉˴˿̈˸̆ʼˍ˄ˁʳʳ̉˴̅ʳ˶̂̈́̇ː˃ˎ˅ˁʳʳ˹˹̂̅ʻ˸˴˶˻ʳ̉˴˿̈˸ʳ˼́ʳ˼́˲̉˴˿̈˸̆ʼˆˁʳʳʳʳ˶̂̈́̇ʾː̉˴˿̈˸ˎˇˁʳʳ˼˼˹ʻ˶̂̈́̇ˑː̀˼́˼̀̈̀̆̈̃̃̂̅̇ʼˈˁʳʳʳʳ̂̂̈̇̃̈̇ˏ˼́˲˾˸̌ʿʳ˶̂̈́̇ˑˎ̂̈̇̃̈̇ˍʳ˹̅˸̄̈˸́̇ʳ̆˸̄̈˸́̇˼˴˿ʳ̃˴̇̇˸̅́̆ʳ̊˼̇˻ʳ̇˻˸˼̅̆̈̃̃̂̅̇̆˖˴́˷˼˷˴̇˸̆˖̂̀̃̈̇˼́˺˝̂˵ˍ˦̈̃̃̂̅̇̆˔̆̆˸̀˵˿˼́˺˝̂˵ˍ˖˖˥˸˷̈˶˸̅˶̂́˹˼˺̈̅˸ˍ˄ˁʳʳ̉˴̅ʳ̆̇˴̅̇˲̇˼̀˸ʳːʳ˶̈̅̅˸́̇ʳ̇˼̀˸̆̇˴̀̃ʳΩʳˣˢ˜ˎ˅ˁ̉˴̅̀̂ˁ̂̈̇̃̈̇ˏ̆˸̄̈˸́˶˸ˡ̂ʿ˼̇˸̀̆˸̇ʾ̇˼̀˸̆̇˴̀̃ˑˎ˂˂̀̈˿̇˼̃˿˸̂̈̇̃̈̇̇̂̂̈̇̃̈̇̆̈̀̀˴̅̌̅˸˷̈˶˸ʻ˼́˲˾˸̌ʿ˼́˲̉˴˿̈˸̆ʼˍˆˁ̉˴̅˼́̃̈̇ˎ˂˂̈̆˸˷̇̂̆̇̂̅˸˼́̃̈̇˷˴̇˴ˇˁʳʳ̉˴̅ʳ˶˴́˷˲̆˸̇ˎ˂˂ʳ̈̆˸˷̇̂̆̇̂̅˸ʳ˶˴́˷˼˷˴̇˸ʳ̆˸̇̆̈̀̀˴̅̌˴̇̇˻˸̃̅˸̉˼̂̈̆̇˼̀˸̆̇˴̀̃ˈˁ̉˴̅̅˸̆̈˿̇ˏ˼̇˸̀̆˸̇ˑˎ˂˂̈̆˸˷̇̂̆̇̂̅˸˷˼̆̇˼́˶̇̅˸̆̈˿̇̆ˉˁ˹̂̅ʻ˸˴˶˻̉˴˿̈˸˼́˼́˲̉˴˿̈˸̆ʼ̎ˊˁ˼˹ʻ̉˴˿̈˸˼̆˴̆̈̀̀˴̅̌ʼ̎ˋˁ˶˴́˷˲̆˸̇ˁ̃̈̇ʻ̉˴˿̈˸ˁ̇˼̀˸ʿ̉˴˿̈˸ˁ˼̇˸̀̆˸̇ʼˎˌˁ˼˹ʻ̉˴˿̈˸ˁ̇˼̀˸ˑ̆̇˴̅̇˲̇˼̀˸ʼ˄˃ˁ̀̂ˁ̂̈̇̃̈̇ˏ˼́˲˾˸̌ʿ˶˴́˷˼˷˴̇˸ʾ˶˴́˷˼˷˴̇˸ˁ̇˼̀˸ˑˎ˄˄ˁ˼˹ʻ̉˴˿̈˸ˁ˼̇˸̀̆˸̇ˁ̆˼̍˸ˑ˄ʼ˄˅ˁ̅˸̆̈˿̇ˁ̃̈̇ʻ̉˴˿̈˸ˁ˼̇˸̀̆˸̇ʼˎ˄ˆˁ̐˸˸˿̆˸˂˂˼̇˼̆˼́̃̈̇˷˴̇˴˄ˇˁʳʳ˼́̃̈̇ː˺˸̇̇˻˸̂́˿̌̂́˸˷˴̇˴˴̇˶̈̅̅˸́̇̇˼̀˸̆̇˴̀̃ˎ˄ˈˁ̐˸˸́˷˹̂̅˄ˉˁ˹˹̂̅ʻ˸˴˶˻ʳ˶̂̀˵˼́˴̇˼̂́ʳ̂˹ʳ˼̇˸̀̆ʳ˼́ʳ˼́̃̈̇ˁ˼̇˸̀̆˸̇ʼ̎˄ˊˁʳ˹̂̅ʻ˸˴˶˻ʳ˶˴́˷˼˷˴̇˸ʳ˼́ʳ˶˴́˷˲̆˸̇ʼ̎˄ˋˁʳʳʳ̉˴̅ʳ́˸̊˲˶˴́˷ʳːʳ˴̃̃˸́˷ʳ˶̂̀˵˼́˴̇˼̂́ʳ̇̂ʳ˶˴́˷˼˷˴̇˸ˎFig.2.Candidates Computing Job and Supports Assembling Joba candidate sequential pattern.In lines13to14,if the value is the arriving itemset of a sequence,CCReducer stores the input itemset for the generation of new candidate itemsets in the following lines.It is noted that there is only one newly arriving itemset of a specific sequence number at a timestamp.CCReducer has to compute all combinations of items in the arriving itemset in order to generate the complete set of different sequential patterns.For example, if the incoming itemset is(ABC),all combinations for generating candidate sequential patterns are A,B,C,(AB),(AC),(BC),and(ABC).In lines17to 22,CCReducerfirst appends each combination to each candidate itemset in the cand set summary to form new candidate sequential patterns.Then,the newly generated candidate sequential pattern is put into the result set as an output in line19.Meanwhile,if the timestamp of the candidate sequential pattern is bigger than the start time of the current POI,the newly generated candidate itemset is put in the summary of the current timestamp for further computation32J.-W.Huang,S.-C.Lin,and M.-S.Chenat the next timestamp in lines20to21.Note that the candidate itemsets whosetimestamps equal to the start time are not stored.In other words,the obsoletedata at the next timestamp are pruned away.In addition to the newly generated candidates,CCReducer stores each combination with the current timestamp inthe summary at the current timestamp in line23.The summary of the cur-rent timestamp will be used to compute candidate sequential patterns at the next timestamp.Finally,all candidate itemsets in the result set are output as <candidate itemset,null>pairs in lines25to26.After the collection of output pairs of each CCReducer,the candidate computing job has dealt with all incom-ing itemsets at the current timestamp,generated candidate sequential patternsof all sequences in the current POI,and updated candidate set summaries of all sequences for the computation at the next timestamp.3.2Support Assembling JobAs shown in Figure2,the support assembling job calculate supports for each can-didate sequential patterns.The support assembling job reads all candidate sequen-tial patterns from the outputs of the candidate computing job.SAMapper utilizes a local map to aggregate occurrence frequencies of different candidate sequential patterns locally in lines2to6and outputs<candidate sequential pattern,its local supports>pairs in lines7to8.Pairs with the same candidate sequential pattern are sent to the same SAReducer.In lines2to3of SAReducer,SAReducer accumu-lates supports of the same candidate sequential pattern again and gathers thefinal supports.For those candidate sequential patterns whose supports are no less than the minimum support threshold,SAReducer reports them as frequent sequential patterns in the current POI in lines4to5.Then,DPSP algorithm moves to the next timestamp and repeats these Map/Reduce jobs.4Performance Evaluation4.1Experimental DesignsTo assess the performance of DPSP,we conduct several experiments to evaluate the performance and the effects of input parameters.DPSP is implemented in Java and runs on top of Hadoop version0.19.1.Hadoop cluster consists of13nodes and each node contains2intel Xeon(TM)CPU3.20GHz,2GB RAM and32GB SCSI hard disk.The synthetic datasets are generated the same as[7].In our experiments, every point in thefigures is the total execution time of40timestamps and the POI is set as10timestamps unless specified otherwise.The minimum support threshold is set to0.02and there are10000different items in the synthetic datasets.4.2Experimental ResultsFirst,we examine the performance of DPSP with large numbers of sequences as shown in Figure3.Note that both X-axis and Y-axis are in log scale in(a). The total execution time does not vary a lot when the number of sequences isDPSP:Distributed Progressive Sequential Pattern Mining on the Cloud33Fig.3.Experimentssmaller than500k.The reason is that most of the execution time comes from the overhead of Hadoop scheme such as disk I/O and communication costs. When the number of sequences is bigger than500k,the total execution time increases linearly.We show the linear part of Figure3(a)in more details in Figure3(b).The linear equation of the regression line is y=0.0005x+1057.8, which means DPSP possesses very good scalability.Therefore,DPSP shows great practicability with large number of sequences.In the second experiment,we demonstrate the effect of increasing the length of POI.As shown in Figure3(c), the total execution time goes up very quickly.The reason is that the number of candidate sequential patterns generated by each sequence grows exponentially as the length of POI increases.Therefore,the processing time of DPSP increases accordingly.The distributed nature of DPSP helps a little.Finally,we show the advantages of the distributed scheme of our proposed al-gorithm DPSP.The datasets contain1000k to10000k sequences.As shown in Figure3(d),the total execution time drops as the number of nodes increases from 1to8.This shows the merits of the distributed scheme.It is noted that both X-axis and Y-axis are in log scale.However,the overheads of disk I/O and message com-munication retard the reduction rate of the total execution time when the number of nodes equals to13.Nevertheless,the decrease of the total execution time is re-markable.It is still worth to include more computing nodes in the cluster if we want to deal with more sequences.By utilizing Hadoop platform,it is extremely easy to extend the scale of the cluster to acquire better performance.5ConclusionsWe proposed a distributed algorithm DPSP to address the inevitable scalability problem of the progressive sequential pattern mining.DPSP is running on top34J.-W.Huang,S.-C.Lin,and M.-S.Chenof Hadoop.We designed two Map/Reduce jobs in DPSP to efficiently compute candidate sequential patterns,update summaries of sequences,and assemble supports of candidate sequential patterns within each POI.As such,DPSP is able to report the most up-to-date sequential patterns.The experimental results show that DPSP possesses great scalability and thus increases practicability when the number of sequences become larger.In addition,by utilizing Hadoop platform,it is easy to increase the number of computing nodes in the cluster to acquire better performance.References1.Agrawal,R.,Srikant,R.:Mining sequential patterns.In:Proc.of Intl.Conf.onData Engineering,February1995,pp.3–14(1995)2.Aseervatham,S.,Osmani,A.,Viennet,E.:bitspade:A lattice-based sequentialpattern mining algorithm using bitmap representation.In:Proc.of Intl.Conf.on Data Mining(2006)3.Cheng,H.,Tan,P.-N.,Sticklen,J.,Punch,W.F.:Recommendation via query cen-tered random walk on k-partite graph.In:Proc.of Intl.Conf.on Data Mining, pp.457–462(2007)4.Chilson,J.,Ng,R.,Wagner,A.,Zamar,R.:Parallel computation of high dimen-sional robust correlation and covariance matrices.In:Proc.of Intl.Conf.on Knowl-edge Discovery and Data Mining,August2004,pp.533–538(2004)5.Dean,J.,Ghemawat,S.:Mapreduce:Simplified dataprocessing on large clusters.In:Symp.on Operating System Design and Implementation(2004)6.Hadoop,7.Huang,J.-W.,Tseng,C.-Y.,Ou,J.-C.,Chen,M.-S.:A general model for sequentialpattern mining with a progressive database.IEEE Trans.on Knowledge and Data Engineering20(9),1153–1167(2008)8.Kargupta,H.,Das,K.,Liu,K.:Multi-party,privacy-preserving distributed datamining using a game theoretic framework.In:Proc.of European Conf.on Principles and Practice of Knowledge Discovery in Databases,pp.523–531(2007)9.Luo,P.,Xiong,H.,Lu,K.,Shi,Z.:Distributed classification in peer-to-peernetworks.In:Proc.of Intl.Conf.on Knowledge Discovery and Data Mining, pp.968–976(2007)10.Nguyen,S.,Sun,X.,Orlowska,M.:Improvements of incspan:Incremental miningof sequential patterns in large database.In:Ho,T.-B.,Cheung,D.,Liu,H.(eds.) PAKDD2005.LNCS(LNAI),vol.3518,pp.442–451.Springer,Heidelberg(2005) 11.Wolff,R.,Bhaduri,K.,Kargupta,H.:A generic local algorithm for mining datastreams in large distributed systems.IEEE Trans.on Knowledge and Data Engi-neering21(4),465–478(2009)12.Xu,X.,Yuruk,N.,Feng,Z.,Schweiger,T.A.J.:Scan:A structural clustering al-gorithm for networks.In:Proc.of Intl.Conf.on Knowledge Discovery and Data Mining,pp.824–833(2007)。