Changeset 448

Show
Ignore:
Timestamp:
07/21/07 23:07:35 (1 year ago)
Author:
cdavid
Message:

Change examples and tests following the change of package name pyem->em

Files:

Legend:

Unmodified
Added
Removed
Modified
Copied
Moved

  • trunk/learn/scikits/learn/machine/em/Changelog

    r447 r448  
     1pyem (0.5.7dev) Sun, 22 Jul 2007 11:05:58 +0900 
     2 
     3        * pyem is now part of the scikits.learn package 
     4        * Renamed to em 
     5 
    16pyem (0.5.7dev) Mon, 28 May 2007 11:31:08 +0900 
    27 

  • trunk/learn/scikits/learn/machine/em/README

    r447 r448  
    1 Last Change: Sat Jun 09 12:00 PM 2007 J 
     1Last Change: Sun Jul 22 12:00 PM 2007 J 
    22 
    3 pyem is a python module build upon numpy and scipy 
    4 (see http://www.scipy.org/) for learning mixtures models 
    5 using Expectation Maximization. For now, only Gaussian 
    6 Mixture Models are implemented. Included features: 
     3em is a python module build upon numpy and scipy (see http://www.scipy.org/) 
     4for learning mixtures models using Expectation Maximization. For now, only 
     5Gaussian Mixture Models are implemented. Included features: 
    76     
    87 * computation of Gaussian pdf for multi-variate Gaussian 
    …… 
    1110 * Confidence ellipsoides with probability at arbitrary level 
    1211 * Classic EM for Gaussian Mixture Models 
    13  * K-mean based and random initialization for EM available 
     12 * Experimental regularized (MAP) EM  

  • trunk/learn/scikits/learn/machine/em/doc

    • Property svn:ignore changed from
      *.pyc
      .*.swp
      *.so
      *.pyd
       to
      *.pyc
      .*.swp
      *.so
      *.pyd
      *.out
      *.tex
      *.log
      *.aux

  • trunk/learn/scikits/learn/machine/em/doc/Makefile

    r363 r448  
    1 # Last Change: Mon Jul 02 09:00 PM 2007 J 
     1# Last Change: Sun Jul 22 11:00 AM 2007 J 
    22 
    33# This makefile is used to build the pdf from the rest file and inlined code 
    44# from python examples 
    55 
    6 py2tex  = PYTHONPATH=/home/david/local/lib/python2.4/site-packages pygmentize -l python -f tex 
     6py2tex  = pygmentize -l python -f tex 
    77rst2tex = PYTHONPATH=/home/david/local/lib/python2.4/site-packages rst2newlatex.py \ 
    88                  --stylesheet-path base.tex --user-stylesheet user.tex  

  • trunk/learn/scikits/learn/machine/em/doc/index.txt

    r363 r448  
    11.. 
    22    restindex 
    3         page-title: Pyem 
    4         crumb: Pyem 
    5         link-title: Pyem 
     3        page-title: Em 
     4        crumb: Em 
     5        link-title: Em 
    66        encoding: utf-8 
    77        output-encoding: None  
    8         tags: python,pyem,Expectation Maximization,EM,online EM,recursive EM 
     8        tags: python,em,Expectation Maximization,EM,online EM,recursive EM 
    99        file: basic_example1.py 
    1010        file: basic_example2.py 
    …… 
    1414    /restindex 
    1515 
    16 .. Last Change: Mon Jul 02 09:00 PM 2007 J 
     16.. Last Change: Sun Jul 22 11:00 AM 2007 J 
    1717 
    1818=================================================== 
    19  PyEM, a python package for Gaussian mixture models 
     19 em, a python package for Gaussian mixture models 
    2020=================================================== 
    2121 
    2222.. contents:: Tables of contents 
    2323 
    24 PyEM is a package which enables to create Gaussian Mixture Models 
     24EM is a package which enables to create Gaussian Mixture Models 
    2525(diagonal and full covariance matrices supported), to sample them,  
    2626and to estimate them from data using Expectation Maximization algorithm. 
    …… 
    3030online EM (ie recursive EM) and variational Bayes implementation. 
    3131 
    32 PyEM is implemented in python, and uses the excellent numpy and scipy 
     32EM is implemented in python, and uses the excellent numpy and scipy 
    3333packages. Numpy is a python packages which gives python a fast  
    3434multi-dimensional array capabilities (ala matlab and the likes); scipy 
    …… 
    4141 
    4242.. _scipy: http://www.scipy.org 
    43  
    44 Pyem depends on several packages to work: 
     43.. _scikits: https://projects.scipy.org/scipy/scikits/ 
     44 
     45The toolbox depends on several packages to work: 
    4546 
    4647 - numpy 
    47  - matplotlib (if you wish to use the plotting facilities of pyem) 
     48 - scipy 
     49 - setuptools 
     50 - matplotlib (if you wish to use the plotting facilities: this is not mandatory) 
    4851 
    4952Those packages are likely to be already installed in a typical numpy/scipy environment. 
    5053 
    51 Since september 2006, pyem is included in the sandbox of `scipy`_.  The sandbox 
    52 contains packages which are pending for approval in main scipy; that means it 
    53 is not installed by default, and that you need to install scipy from sources. 
    54 For the most up-to-date version of pyem, you need to download scipy from 
    55 subversion, which contains the development branch of scipy. 
    56  
    57 To install pyem, you just need to edit (or create if it does not exist) 
    58 the file Lib/sandbox/enabled_packages.txt in scipy sources, and 
    59 add one line with the name of the package (eg pyem). 
    60 After, you just need to install scipy normally as explained  
    61 `here <http://www.scipy.org/Installing_SciPy>`_. 
    62  
    63 You can (and should) also test pyem installation using the following: 
     54Since July 2007, the toolbox is included in the learn scikits (`scikits`_). 
     55There is no official release yet, but you can get the package through svn with 
     56the following command: 
     57 
     58svn co http://svn.scipy.org/svn/scikits/trunk scikits.dev 
     59 
     60Of course, you can also use a graphical tool such as TortoiseSVN on windows if 
     61you do not feel confortable with the command line. Then, you can install it 
     62with the following command in the scikits.dev directory: 
     63 
     64python setup.py install 
     65 
     66You can (and should) also test em installation using the following: 
    6467 
    6568.. raw:: html 
    …… 
    6770    {+mycoloring} 
    6871 
    69     from scipy.sandbox import pyem 
    70     pyem.test() 
     72    from scikits.learn.machine import em 
     73    em.test() 
    7174    {-mycoloring} 
    7275 
    …… 
    7477============ 
    7578 
    76 Once you are inside a python interpreter, you can import pyem 
    77 using the follwing command: 
     79Once you are inside a python interpreter, you can import the package using the 
     80follwing command: 
    7881 
    7982.. raw:: html 
    8083 
    8184    {+mycoloring} 
    82  
    83     from scipy.sandbox import pyem 
     85    from scikits.learn.machine import em 
    8486    {-mycoloring} 
    8587 
    …… 
    8890----------------------------------------- 
    8991 
    90 Importing pyem gives access to 3 classes: GM (for Gausssian Mixture), GMM 
    91 (Gaussian Mixture Model) and EM (for Expectation Maximization). The first 
    92 class GM can be used to create an artificial Gaussian Model, samples it, 
    93 or plot it. The following example show how to create  
    94 a 2 dimension Gaussian Model with 3 components, sample it and plot 
    95 its confidence ellipsoids with matplotlib: 
    96  
    97 .. raw:: html 
    98  
    99     {mycolorize;input/softwares/pyem/basic_example1.py} 
     92Importing the package gives access to 3 classes: GM (for Gausssian Mixture), 
     93GMM (Gaussian Mixture Model) and EM (for Expectation Maximization). The first 
     94class GM can be used to create an artificial Gaussian Model, samples it, or 
     95plot it. The following example show how to create a 2 dimension Gaussian Model 
     96with 3 components, sample it and plot its confidence ellipsoids with 
     97matplotlib: 
     98 
     99.. raw:: html 
     100 
     101    {mycolorize;input/softwares/em/basic_example1.py} 
    100102 
    101103.. raw:: latex 
    …… 
    121123 
    122124If you want to learn a Gaussian mixture from data with EM, you need to use two 
    123 classes from pyem: GMM and EM. You first create a GMM object from a GM 
    124 instance; then you can give the GMM object as a parameter to the EM class to 
    125 compute iterations of EM; once the EM has finished the computation, the GM 
    126 instance of GMM contains the computed parameters. 
    127  
    128 .. raw:: html 
    129  
    130     {mycolorize;input/softwares/pyem/basic_example2.py} 
     125classes from em: GMM and EM. You first create a GMM object from a GM instance; 
     126then you can give the GMM object as a parameter to the EM class to compute 
     127iterations of EM; once the EM has finished the computation, the GM instance of 
     128GMM contains the computed parameters. 
     129 
     130.. raw:: html 
     131 
     132    {mycolorize;input/softwares/em/basic_example2.py} 
    131133 
    132134.. raw:: latex 
    …… 
    147149------------------------------------------------------- 
    148150 
    149 The GMM class is also able to compute the bayesian information criterion 
    150 (BIC), which can be used to assess the number of clusters into the data. 
    151 It was first suggested by Schwarz (see bibliography), who gave a Bayesian 
    152 argument for adopting the BIC. The BIC is derived from an approximation 
    153 of the integrated likelihood of the model, based on regularity assumptions. 
    154 The following code generates an artificial mixture of 4 clusters, runs 
    155 EM with models of 1 to 6 clusters, and prints which number of clusters 
    156 is the most likely from the BIC: 
    157  
    158 .. raw:: html 
    159  
    160     {mycolorize;input/softwares/pyem/basic_example3.py} 
     151The GMM class is also able to compute the bayesian information criterion (BIC), 
     152which can be used to assess the number of clusters into the data.  It was first 
     153suggested by Schwarz (see bibliography), who gave a Bayesian argument for 
     154adopting the BIC. The BIC is derived from an approximation of the integrated 
     155likelihood of the model, based on regularity assumptions.  The following code 
     156generates an artificial mixture of 4 clusters, runs EM with models of 1 to 6 
     157clusters, and prints which number of clusters is the most likely from the BIC: 
     158 
     159.. raw:: html 
     160 
     161    {mycolorize;input/softwares/em/basic_example3.py} 
    161162 
    162163.. raw:: latex 
    …… 
    170171    :height: 400 
    171172 
    172 The above example also shows that you can control the plotting 
    173 parameters by using returned handles from plot methods. This can be 
    174 useful for complex drawing. 
     173The above example also shows that you can control the plotting parameters by 
     174using returned handles from plot methods. This can be useful for complex 
     175drawing. 
    175176 
    176177Examples  
    …… 
    202203TODO (this is fundamentally the same than pdf estimation, though) 
    203204 
    204 Using PyEM for supervised learning 
    205 ---------------------------------- 
     205supervised learning (e.g discriminative analysis) 
     206------------------------------------------------- 
    206207 
    207208The following example shows how to do classification using discriminative 
    …… 
    228229==================== 
    229230 
    230 Pyem is implemented in python (100% of the code has a python implementation), 
    231 but thanks to the Moore Law, it is reasonably fast so that it can be used for 
    232 other problems than toys problem. On my computer (linux on bi xeon 3.2 Ghz, 2Gb 
    233 RAM), running 10 iterations of EM algorithm on 100 000 samples of dimension 15, 
    234 for a diagonal model with 30 components, takes around 1 minute and 15 seconds: 
    235 this makes the implementation usable for moderately complex problems such as 
    236 speaker recognition using MFCC. If this is too slow, there is a C 
    237 implementation for Gaussian densities which can be enabled by commenting out 
    238 one line in pyem/gmm_em.py, which should gives a speed up of a factor 2 at 
    239 least; this has not been tested much, though, so beware. 
    240  
    241 Also, increasing the number of components and/or dimension is much more 
    242 expensive than increasing the number of samples: a 5 dimension model of 100 
    243 components will be much slower to estimate with 500 samples than a 5 dimension 
    244 10 components with 5000 samples. This is because loops on dimension/components 
    245 are in python, whereas loops on samples are in C (through numpy). I don't 
     231The package is implemented in python (100% of the code has a python 
     232implementation, so that it can be used for educational purpose too), but thanks 
     233to the Moore Law, it is reasonably fast so that it can be used for other 
     234problems than toys problem. On my computer (linux on bi xeon 3.2 Ghz, 2Gb RAM), 
     235running 10 iterations of EM algorithm on 100 000 samples of dimension 15, for a 
     236diagonal model with 30 components, takes around 1 minute and 15 seconds: this 
     237makes the implementation usable for moderately complex problems such as speaker 
     238recognition using MFCC. If this is too slow, there is a C implementation for 
     239Gaussian densities which can be enabled by commenting out one line in the file 
     240gmm_em.py, which should gives a speed up of a factor 2 at least; this has not 
     241been tested much, though, so beware. 
     242 
     243Also, increasing the number of components or dimension is much more expensive 
     244than increasing the number of samples: a 5 dimension model of 100 components 
     245will be much slower to estimate with 500 samples than a 5 dimension 10 
     246components with 5000 samples. This is because loops on dimension or components 
     247are in python, whereas loops on samples are in C (through numpy).  I don't 
    246248think there is an easy fix to this problem.  
    247249 
    248 Full covariances will be slow, because you cannot avoid nested loop 
    249 in python this case without insane amount of memory. A C implementation 
    250 may be implemented, but this is not my top priority; most of the time, you 
    251 should avoid full covariance models anyway. 
     250Full covariances will be slow, because you cannot avoid nested loop in python 
     251this case without insane amount of memory. A C implementation may be 
     252implemented, but this is not my top priority; most of the time, you should 
     253avoid full covariance models if possible. 
    252254 
    253255TODO 
    …… 
    267269 
    268270 - add other models (mixtures of multinomial: easy, simple HMM: easy, other ?) 
    269  - add bayes prior using MCMC (hard, use PyMCMC for sampling ?) 
     271 - add bayes prior using MCMC (hard, use PyMC for sampling ?) 
    270272 
    271273Bibliography 

  • trunk/learn/scikits/learn/machine/em/doc/tutorial.pdf

    r363 r448  
    5959endobj 
    606044 0 obj 
    61 (Using PyEM for supervised learning) 
     61(supervised learning \(e.g discriminative analysis\)) 
    6262endobj 
    636345 0 obj 
    …… 
    8383endobj 
    848460 0 obj << 
    85 /Length 1902       
     85/Length 1925       
    8686/Filter /FlateDecode 
    8787>> 
    8888stream 
    89 xÚÕYKsÛ6ŸûWðḦ́,|Ž§ŠM3 
    90 ÜSÓMS²ÆŠše'î¯ïîbAB"‰Œ›&NFE,vûíAÿDP”±,EçEœH 
    91 ÔíEl`åí 
    92 ` 
    93 %²8+SxžYŒT¢â2ɃH«8Ï 
    94 —fB+'9Ðæ2¥Î‘øõõÅw¿€*°œÈƒëu ’,Ö© ,Ñõíáûç•ß\œZEi’„•ùÚ¯d 
    95 âÂñþëà³³ hDX¯dÞ¯tJ¿6ði@„kŠ?˜Ÿoqù>}¿å­»ÕŸ×¿BšXš4ˆ€Œ‹\*-’|B©žãÀ 
    96 ÛnI 
    97 Þ¢Œä„ûá`e,R•áÁ"‘Š8-œ_ÄynNvœ*z7ž­é]®H¹µy¬ÍÁRˆÃIžxs=˜—‘Ô%˜MÉY°˜$rhÈþJÕÆÕ<ãƒJ^¢ 
     89xÚÅYKsÛ6ŸûWðḦ́,|Ž§ŠM3‹{jz )ZÖØ5¢ìØýõÝ],HH$‘qÓ$“QD‹ÝÅî·/X      üAQƲAžq"uP·I°•÷‚)”Èâ¬Láyf1J‹<VI 
     90DZÅyVž4Z!ŠXç@›Ë4–:Gâ·×?ü–ª@Âö"®oI 
     91*1Ñõú¯°i߬¢4IÂÊ|íW2_V"<ÞÁ 
     92|vvAë•ÌÂû•Né×> 
     93ˆð–éæç{\~„Oßoyënõ÷õïA ‡V*ˆ€Œ‹\-R<£PÜp`†m·Š€×(âáv8R§*Ã#E"qZ»\ÄynÎtœ*v7ž­é]®H·[óX›s¥       †'’pñîz0,ûP—`0%gÝÄ$‘CC–‡#Š$.“Õ<ãƒJ^¢ 
    9894=ž¹B-ñS 
    9995Ñ 
    100 Z\f`q®ŽNt\Âa}úX׊}¥K¿ÎY 
    101 ’úufÔùŠB 
    102 ·äQšéðWLª, 
    103 å¢Æ£€‰6–f¢ŽÐŽ:BèXåi µŽ%87êóÓᫎ¬Ô=ý9•a_µ{6밄+àø|kš,I‡,Ž3ŽækôOvêƒñ{©U¯"U–/ý*þáŸé×Äà        Dú`¥y{IäÐLÍ}Æ­œ ®W˜×'mW‰8+4‰{]õÖӂŠ·ðŽÞ)€b·6ßK0á~Ê_ø»j 
    104 úÅ;­­™ŽkÍï[ŠEÊZQ‘诜?6„¿‘f?Ç j?°ÓtRøӉÈc)s:aTèÇÛ§jW#ŽY 
    105 »œÜ@“ˆL„Ù¢Ö£4×©4ïuIgyÁ^g* ,Àåšr±“]îV\ä³NH¹hËÎ× 
    106 ÎVÝ€S=2U7ra/¯lÕ<á©ÌõWó76×ßFšesL¹ŒœGØKʗÊáA”^³4šÐLU ÿï¹1À†DÁ 
    107 Ó‹!M±€Ž#lÙÝÎ5šu7•i8>Iûœ?+8šâ•qö5®SŽ]·ÜÁÀ‹¥t˜—âßNPÿU   c3ù<r$YtHÇØËÐú$Yd=¢`µŒ 
    108 /BÖñ:Ë¿œßcUj[ô`²ÙTÀ4‘C45þ9'4ŸH¡öȱöÉÀN‹Xçr 
    109 ì÷ðHãÕ<è=nbšbÔû†K¥'jWv§‘ÉÿŸl%?Ÿ#‘_ÇÞ 
    110 |}ò^RZpJþÉÈÒ F¿uÇÆô0Tñá{ 
    111  À»îVð}h©•šy 
    112 •Y)“P,vdŽø/êÈRxÈŽö 
    113 ÃÒÐäüîçwŠ(&Ê)ŠB/)êø2E!œŠÒß:ZjÓ¶vŒ§Ê·9Tû»g°žP¢ô©;ŠYÿs]NÃʊ R¥â€4Éæ,Ö%DèÖ4°2¥BŠ|}Rcky¿Ò™™qùãÝÐÂڝÙÞ`°W7<›’¿ êÍ®zœ{³øw8×-;óúŠ¹ÏÍ®ŒŒ,s÷’ 
    114 V>$:¡ŽÕì¯6 
    115 éô`Vm7‹Ê®™­¹Ä@߉ØZ‘Hcö«ºÃã=ÑÑm‰#&5g¶‹±K5ÝÙÀkN”6YRë}4ƒ@ 
    116 |ÑÀ?˜ÙŸÙã"Zºi™zìǝbÒVM¶› 
    117 §£8ñ71fm4JbÌ¡œNt¬õÑ2f†,%tªOVüﯠ  
    118 Ý 
    119 m:wö@EÛ—ah9²ºré{<mŠHm‰œæ]M@H 7™s5L°Ú<8w*{bµµtœ‘0Ô9xn‘±u«z4c3Ï°ŠÏ3s     èúç+ó~f¢º#ŒíXE;ó4³a~°ƒÝÞÁ6z;ØmUêûŠïg 
    120 ïP,êG¹±•js•‘̵ 
    121  
    122 [Ÿ“eÇœœtôÂج_2]ešŠ 
    123 7ƒ ?žgŽKÏn€ÙCÇ8àH©nÇ껚 
    124 .1~I~ä^¿1¡%’Ìv\$-j«ÁôDˆ-ÄЕM“!{Æ£6“ƐæÄ%LãF¡6$¶!àŽ±¹Êžä¹óŸ¯P¶ØÇYæd3&ؙÇ=®=;!Mr Â¹ÂBrR¿Ð' 
    125 kÍ'ŠÝHk£K²(z;rNß6Q>®N«XOS>ãóÛ     s9š±,‚Ÿs§„ 
    126 tN’Ù1-˜@œ 
    127 ¥ÿ cŒî_AlÑ»ù4ek_K°?ñ 
    128 :ÂÁMŸÏ†U]ímíÞ:þ?—ó¬ Ój 
    129 H•ye’уáoF|$dWl$¶²î 
    130 ž:þ`öb‰oÈ!ŒÍæŸõ˜:’Ûîn8eز: 
    131 [²ò ntðíXSjÃd=t0CöégŒt25!—À{[‡©F@×Пÿà$UÌÈÍæÆâgÿdpçC'P÷ŒN÷ÆuLá/Š#w©:gYo·ëÐ Ýîð'-&E«LöoEóZØÝ>- 
    132 šiŸkùd0ˆðu­Inè0Î=yÃbÇmçQ{ïÙ».æ 6‰mÜ4”—ºXŠ‚|Ÿï—N    Fž(ïH‡ÓŽŽÓÓÊRÅ¥Òг4Dx€þ 
    133 ªU™$ 
     96\f`q®ŽNt\Âa}úXPEàl¥K¿ÎYØHý:3 
     97ê|S¡Â[BA”f:|ÄßÕ 
     98&U–†rQãQÒDK3QGhG 
     99!t¬ò4ÐZÇ`úür@÷UGVj‡€D8§2ì«vÏf 
     100–p€ÏkCeI:dqœ¡5_#>Ó{©U¯"U–¯ý*þãŸé×ÄàI¬G+ÍۛH"‡fjî3>híeçz 
     101Yçú€ 
     102ÎU"Π
     103MâÞVœEž é­{ÚÁß)àÅîÖ|/¹        ÷SúÂßUÛPÐ(Þií–I»Öü^S,RҊŠD%Çùýdžðúo€YöŸcPµè?°ÓtRøӉÈc)s:aTèçõSµ«ÑŽY 
     104»œq¹qM"2f‹ZÒ<š;Si 
     105uIgyÁš3P9*\ ²Ë݊kÜa„”‹¶ Ÿæp¶ê&ê‘©º‘ £Œ~°Eó„ g€2×_ 
     106ol/ÞFšeŒ9Š\öœGØkʗÊáA”^ŒYTè 
     107Š‡ª 
     108ÿ÷Ü`?¢`éŠXRÚ¶ ·sfáŠ2 
     109Ç'iögUŒ20a¬qŠ ëZs/–Òa^Šÿ;A}¯Æfò!r$Y€cìe×ú$YÏzD 
     110ŽÕ2RŒÊ³þˆ×Yþ­Üùm}¬J 
     111'¹ 
     112L6› 
     113˜&rˆŠÆ?ç„ÆÉ¢«=òF_ûä 
     114ÎNqä2œ|41J 
     115|b/ö 
     116§|J7Ô~ìŠMæÇD'ž 
     117¯Ö–AmÑÚMCž?QrzZš#pt#èõÀV¬"‘$ß×Ál&¿ƒG"ƒ 
     118ƒ{ 
     119ì“÷šÚ3ŠþÑÈÒ FtÇÆ41Tòá{ 
     120 àá¶[Á÷¡¥^¢æTf¥LB±Ø’9â¿š%Ká!ÓÚ{ KC£ó‡_?˜ª˜(§* 
     121œ€š#àË 
     122ü*JïhišOÛÚùžJßæPíï^ÀâB‰Ò§î(f9þÏu9š+2H•Š“R mfv)MÌÒ0L££ä¹£Æžò~¥3;+§á§»¡óƒµ;³œ¡0Ÿá¡ŽaNÔ*š]õ8ð2Ÿ÷žÃ¹fÙNWÌ}n‚]W†%HvnWH'#›€ªMÇ© 
     123Øm¬4Uì‘U} ÌDl¥H€1ã©îðxOtt›ÊˆIM™Ë¶/v©Š3ÃkΚ6«RÏ}4“‚)³‰&ýd0<:C?³ÇEŽŽ¹æ’±wvŠI?5ÙnvœNŠâdRÄߏ6ûšôŽFIŒ9”Ó‚ŽÅ¢>ºÝdãêÙMüÿLh(Ùo:wè@EÛ—aZ9²ºré{<mŠHmš 
     124Ñ9ŸÌ»š 
     125ý1œdÎÕ n‚ÕæÁ¹LÙ«­¥ë„¡«ç[pXÕ£{œ!Ú>ÏÌퟋÏ7æýà@†êŽümç)ڙN3æ':Øíè`£g¢ƒÝ<šRß7}?+|‡bQ?²ÈmŠ†0wÉ\?È!±å»Xî݈Ò 
     126±Y¿dºÊt;î>?žgŽKÏn~¡c 
     127p€Të1€ú.ª‡Ûˆ_ŽÜ{7æ14À‰M2Ûq‘Žš­>œÓßÌÇЕM“ézQì6Nc@s  ÓAQš 
     128É£mÈqÇ  ês‚ižãéBÙ  
     129œÌme   vfÛ 
     130×^œP&yoÌ¢= î!p4,ÁAþlž)f—°@â1B‘Eю‘³‚nÓD·]U3Õ«§©žýòÇ  s9š+üÒ³§„ 
     131¹:ٛ 
     132“B‰Ô› ýg³`„}1Eïæӓ­y-eŸþ5t„ƒ›6_ «ºÚۚœup?—ë¬ Óÿi@•ye’ЃáoFÏKȪŽ³‘Øʺ7vàªø“ÙwêK|C€06;˜[{Öc 
     133 $çG‚Nö€¬dߖ¬<ˆŸ 
     134kIm˜ÜËuú#¹³1¹`ãîÞÖ_ª 
     135Ð-ôç÷þ'‰§bFŠX67Ööo'ñ=tuÏëtQ\Çöb:âpW*¡S– 
     136ðv·ÍÐÝÃb"Q±Êôgÿ84¯ 
     137ÝíÓ¡™öµ–OƒßÏ^ߍ÷çÇÁê 
     138jc*5wÏÆXTš¹hìž5àÜj/A{~ŽG'qïÄÇOžLEú 
     139>?.YŠ<QÞ¿“–i§–¥ŠK12€hi ðHý\ž& 
    134140endstream 
    135141endobj 
    …… 
    139145/Resources 59 0 R 
    140146/MediaBox [0 0 595.276 841.89] 
    141 /Parent 120 0 R 
     147/Parent 121 0 R 
    142148/Annots [ 72 0 R 76 0 R 78 0 R 79 0 R 81 0 R 85 0 R 87 0 R 88 0 R 90 0 R 91 0 R 93 0 R 94 0 R 96 0 R 97 0 R 99 0 R 100 0 R 102 0 R 103 0 R 105 0 R 106 0 R 108 0 R 109 0 R 111 0 R 112 0 R 114 0 R 115 0 R 117 0 R ] 
    143149>> endobj 
    …… 
    271277/Type /Annot 
    272278/Border[0 0 0]/H/I/C[1 0 0] 
    273 /Rect [114.477 435.459 289.583 447.139] 
     279/Rect [114.477 434.848 344.76 447.646] 
    274280/Subtype /Link 
    275 /A << /S /GoTo /D (using-pyem-for-supervised-learning) >> 
     281/A << /S /GoTo /D (supervised-learning-e-g-discriminative-analysis) >> 
    276282>> endobj 
    277283106 0 obj << 
    …… 
    394400/D [58 0 R /XYZ 89.291 177.676 null] 
    395401>> endobj 
     402120 0 obj << 
     403/D [58 0 R /XYZ 89.291 177.676 null] 
     404>> endobj 
    39640559 0 obj << 
    397406/Font << /F34 66 0 R /F39 70 0 R /F44 75 0 R /F15 84 0 R >> 
    398407/ProcSet [ /PDF /Text ] 
    399408>> endobj 
    400 147 0 obj << 
    401 /Length 2943       
     409148 0 obj << 
     410/Length 2683       
    402411/Filter /FlateDecode 
    403412>> 
    404413stream 
    405 xÚÅZYo 
    406 ¹~ׯh /#@C7Ÿ6ÈC¬7Ùdsx 
    407 ÁzZ3­#ÖL+ 
    408 v~}ê"›}Ž-íÂ0ÆÓMY_ÝEŽt’Â?”•2•NŠ¢T©É’Íî$M®`æ»-Nç*¯,<OL®±JÛdmµ2¥"1Z•ºš^®UZ陝uš©"wɺp¹*œA¢?œŸŒy›ÙD§ªJ«äü2ѺTe™%žè|ûÓêœÑééÏçߟ|{Þq2NYXm6æ(4cŽEÌÑT„+æž?5åêñT¯vwøôiÄ]d)PMG€õ$ ²ÉQIçy9ûŒ&¥ ŒvõxŸ[øŽòpŸ‹Óµµfõ>ÍR¹ä‘Ošö‘_žqâpÍ/žŒÅGKz;4üŒÃ×ð‘× 
    409 ·Ù{Ÿ+ž¿„ÇzsAñD́)ZA"vÿš 
    410 àÔšµ7ou‰Œ‘34žtj-     }Ž`ZDghdwjŠÕ‡Ó,_ÕWÈ'ê{ƌϠ
    411 Éb†oŸp"õ"# 
    412 (Ìä~A4ŽAœõԝ€\F$øi¶ñ,=ÖŒË ÈÂé 
    413 ,"Ÿ#çÔ«7RšŒÓ6 
    414 ‘=Éb„Ó"û 
    415 # 
    416 :Zk«²ŒÔócd’ 
    417 ŠS”`Ê;±C³CU‰¬÷8›¯LšægøXôLÂKÉAx2ìŠR#ø­—>LÒ~‘Ãóz/ŽÌ¶EyžÂ TlÉ«JòÊ©\ 
    418 +8~%âuDMaÓó¡ÁŽš£žª‡ª\•¹í#D«§9Âx°rV,÷yÐO±šŸML]ï#›¯mZš¬0lzWÑî“Á`µ^=_{ÄIŒw†A‚;ö$Ži­)ŠÛ{Yu'MþˆÛ=¯[ž?ÀÇ] 
    419 œôT®W¿åQï*ÇxIS³ˆÄ–“GØVF÷œ|"nS‰³ÁÀ>DDO侔š 
    420 e       ®ÇšŠ•˜ƒžc|ø 
    421 ’§©JbÜ4eËãS ˆfá0réõžã¥‡öQF6ÍAñØÛÓґöc@ 
    422 Ïíڃ {«¬       ÈzË áÅ4n|žá<2Ö 
    423 ååÌö’À¹šhÀ|_|Ñ¢­wˆçœ¯õ–§ú’ãºXr¢x€0SPÉ4žhìÆ°r:Fh*.]°y 
    424 5 œ‹“éX‘ 
    425 "p„Œî XO*0gcIQJ>?--9‡13ÎaôPë8ÔyŒüG’.ÜèÈûŒÞ šm(»4% 
    426 9Œnžªy×D—<-Ûʎ•m€óq‡³ÍGaBbPñ8®»­ß§Úù˜ä‘¿ro‚ån:íÁ e¡ú¢ æA†Î\ŠªÒ$¹üZhLŸÿ=ùéç4ÙBŠþþ$UÎæÉ3Ö 
    427 et™ìN µ‚ 
    428 ‚ßnO~<ùg—Éý^ëh³q&ϲTY—Ž3¹VI"ùHߔ¥]ý4žAìW>+æù0.p$Îg<æ3=o»g 
    429  
    430 ôíŒ-N4ô웯k&£ŠIË;y!—γ!ÍXH 
    431 §j®<Ù “Ã΄xþþÒ7 
    432 ÷Æ0èýz 
    433 r±“c„u©ŠßŸNFÑ($Ã-2ß 
    434 ú 3µô£äÌŸ­ 
    435 )dìwVʕ.É*èà {¬7êuD>Ñ# öD%\{u5#šátÖC0ìM„æßÁNܛË˜Âógl—S“CÊÞ÷Á=QÿԜûªº 
    436 ÉÁúŠúÐv§ŠÆ›¹ßvÎX*šZøè«DtœWšwñpy 
    437  Ÿ*XŽÞY^šŽš&õ+$눆Ôë\¢ 
    438 e       P¯ß'ƒƒ¶h÷ÜÍ/ 
    439 µG|LjšÁà*8‰/C$cû] t‰–ËÐߣSõ 
    440 '£‚RB3h¢æÆ7ù<^óWh⌊ÉaÔbò`ß'6\oøål‚=û”ásĎ»QŠÚ 
    441  
    442 xˆœžzŠŸ× 
    443 ÞžéŸŽ›v·žÈ@»P–æœÃU¥*œ^ŽLDœC« 
    444 û`{UTì 
    445 DÕ@U&ŒWgì ä;ÔbHîWŸ Hžã“øCo’¿v²æãÃãýœ¹Ò(WžeQ:šY/s¹S&-I”?÷­×;KíðˆY™ÕRp[v`šzÅOžCc­W–ééüYÍá 
    446 HWßÁûü()ÉXü,õù‡ù zîöƒWàôu·‘ 
    447 vo>dm*{Ž]i–ÁéÚYޔ;, 
    448 ±Û+qÒçTgÓÕ·Ì 
    449 âùY.`t‚Û< ó€çUûžm‡ÿý/ŠÎTµqO<Å¥ÒÃÝKUÆio|3^û4QïùÁ_ŠŒn'øb     »Ö²ìL›ÛÒ²b3áwìG†Š<Ý¥À.cSк    õžñ‚Cœó՞ 
    450 . 
    451 ×TBA®S–ñ]íéÉO‡S<ì°Tfx1µë    áêI‹!ˆáØ6|÷-<ö5†#5þòǀ]³ŽR0ULi ‡G£Ñ^Yå„ÄŸõqNÍRHûÔ³âY7–”VøÓ|wÄ 
    452 + V8ÖwÏ[:µø*9Á–ùJ=Ã>î6ºtŒ£^Ã0Yߓ¶_Œ?¯2w–òÕÊ 8Yh‹÷æ)u‚ý‘w)5ZO)5·ã”êiüa'@[‰Å‡ñ 
    453 3š±¬zà9=S 
    454 .äj±Ñ 
    455 îÞÈ»Þåt¥lšî©bä{I{ 
    456 ýø>+TYŸú Ž­s«òÔ 
    457 C-T1êÚß²õ™Jmþ«è·LU‘ÙcH 
    458 *Fú·Ê2W._ô‚uU(]”>+¿Ø‰ @– 
    459 qb¡ùr'î6ÿåx    xpâ9;®œ@(5Ì9ñg¢~‘/¢öN 
    460 ¡^vâ_A¿Á‰‘z'Žþc։çQöœØU/÷b0ƒÎuы 
    461 ±†kKòæ!nmøBm:¥Ýü%Ú՝$—{ªñNœ]QüDý‹æÓ 
    462 t([)x­ô’*¢mŠ$5̘•ô% 
    463 oLÎ÷&‹’zªXÒÏÊ0&ä¥YBŸö4G ä¡*{žžélª¬Ž,aö-ïG&=5,g(nÂQ™ü 
    464 ø*Ÿ‘ŸKeË,–~ôÛw×®Ë/ѐÎÁ)šHhXGëÌØ՟ÄÑï 
    465 ßI|ÜË÷NŠã›ŒÃT“šñJ 
    466 kXg,2C߉kðK$N­2úˆÄB#C~B+‘ˆ!øk“VY¯S»èk¢åðq#eðQ„ï+cdz£rLã 
    467 Ê       Û[e‡yüš€ð2‹šð4šØFš'‘m/²ÞHZُäÔZc’Ã8¥ÍèDÁ1E®Žæ›¡sEÆ‘Æ 
    468 êÇÿGëô+       ¯_GLU’>„ò»1.71”¥z÷%(¥<‚Ršb”vÜSrK·€²kÙyó€_ã¥ð/És梗›§Û`Á<”yóDP~qó,¢ôæ‰PšYóÌ£\;ˆ{¡u PŒ"IØ úÊl9IÍ×-¢_¡P;ÌPP 
    469 #%[2tk_Ÿ­V8Þ¢!„&ªÔÿ’Rì+PÃ7;8uz>z}–Ùº~ã 
    470 $Ÿ1‰OïÜÓP×Ó«\¶úzPË^­Py$|i‹ ž¢–á&j¯7ëð÷ŒÒ@š$ÒÇžrTV9£_¯ó±ŒXÖ¹ÐLTÈ觕eMÞwÔ 
    471 ßGmÕ§QqiîéñO±•*­I"¡ÆŠƒŽ3s¯Wœ®œ*u¹š8O3¡ž     ?»éՄÛ{ŠÚDMÕ¶Ó|ˏJ'Žd:ÓD÷œ^»ûþZ 
    472 zžŒñ&œv»“Õ>jF', 
    473 ¿Ã?\íô6N$Ðòë<œmJHZyµl¡™°Íñ€óæ[I&í،‘)_fFúßö-s, 
    474 ±öÐPeŠršŒ‘'ÿÒÑê쵝èŽØIhb;E‘2ê3u1ðíK© 
    475 ·}ªAr»Œþâ–O9Âr Ï9ùŒíŽ9ßÙÈ^Ž?¶J"eNtJ0çà<•ij™"šÙ_x5 
    476 Tә_x¡%€Ò„Í$MöpYŠEå’h§!×ÿÙ‰ƒ 
     414xÚÅZ[oÛ8~ï¯0°/3ŒéV`_vÑéî,fv1 0˜îƒj«IØ,§—ùõ{n”HI–ÛŽƒ¢p-‘‡äw>ž阠
     415†fQVÊVfQ¥Ò6[¬·/ôâz^¿0"áM®òÊÁóDçÊ[§Œ[¬œQ¶tS"ÖšÒTÓÍҕ91³Ñ™*r¿X>W 
     416·(ô·ëW?dnaŽªtµž~·0ŠTe™-‚Ðõæ÷åkôÅÿ®|ñêº_Ézå`H4ÙxE‘¯XÄ+ڊpÅ+î.l¹|º0Ëí#>}­.ºH“­ÎèÛ Íè+Bgõ[±Ówžâ€ŸÑŠíú”=£«o“ŸÑµšÑU„Îê:·b§ëpÅI]£Ûæº>‘Ÿf‰Ïû‹•-–ûxlOj¯µ²ÚÑŸšÑ^„Îj?·b§ýpÅIí£·õQ4Gu÷GyÀí{^ï—otŠ¥å 
     417·|B»Ø?áK¶ü€ 
     418í-÷{؎„¶ 
     419?có-|ä5Yí(sïàsÃýïà±& P‚PÓŸLg€Å¹)zÜñäi[ãÛ§\!µ@ØÊ8•y¶tB|€ÿ>ÁT5(sõƒÉ"ÞVÀ[†ûi­ª²ŒÆ]ãê{TÐ:KÕë °šû‹,_Ö7MË 
     420õÕÆç Ì=Ò×<КС;›àÜæa@24Aœ   Ò=Y-#ü4›ž— 
     421kžåÈ^̛ŒÝkD뎢šY^%žOÓ4$ö^#œ=.¿åv¡&xýoŽ­kÔ'˖?ŠŸÕ3¿ŽZ—ü    
     422 
     423‘‰Ù#ÅY^öe» 
     424 
     425d 
     426M`Däv“KoŽCÒÏÊßGF'  üҕŽ5‹<ÓÊHfÅöÍü9h^õÒ䟉 
     427fìqŠh€ÃÃZç    Ža‘ÙÕóP8BÀý%vÈä›CÏ_Ž;Ù-xÜ¿Ñ&'Ø`gh+g5DیíƒgŠY˜ 
     428ÎqT¡X|‰ 
     429è Œ™G~í£Œ¬Á«éᩑî8ÆÀlcäö FvŒ³ßÜòøö=Û>dßN΍ñi 
     430LòC 
     431Á@bœßöqç%± DÇ>/¹ÞâŸsªÌ³Eìà ҫH|ŒÇÃ9Åmpã^b 
     432žpžû&­JOm5oz¹«       [œ: 
     433ÕO"t?ò"š>œ/S-ædæuC+Ó¥It›€€BxxÏÆiMiØ$óoÎu¶(`_žD 
     434¶Á@U”œ!R?">P$n)bU]òCñš¿npŽú1Ê[ 
     435‡‹2 
     436u4Y‹õÞÊì-O|}Q:fž‘6ký™çÚ œÎ"7Êwœr‰&,HœLØj—R}Iö߀÷ÂKíhBÓê߆€Ä"‰;ù2q§LÇÎBÝQêhOÁ±g'ÁÁG›€/ 
     437 
     438|5™IâNâ<Šhª 
     439ŒÎ:÷8Ï%àÁïz— 
     440Dy[€›(É6ë€49DºÔ 
     441–ÛËòi]«¢¬ÞãL™ŠŽã(ýYÛ0äJAkÝކpú *Qê ŸÎ1BømZ    Ù͖§+š:}A!rªA ý‚Èürzähœ¹\™²˜Œ<"²Šd(ðx¿0Pqû 
     442O˜âOUòyñ-ø«89Ë¡ÆwLB'0ÀYÀTÕ<†^f 
     443üDÈWp–p!ˆ 
     444êËé8êŸ.ÅçŽþ»`ŽØî0pa÷Žé࣒<š.x ¶Ûµ\ŽÖ6bH2”žTØ 
     445ÄAuê7ÚNgr=aÚ 
     446‡&æð¹n,Ÿ6±K>/TYÌ[J$ÓYŠáSC¿K™S•œŽÿŽ”5õ‘ÀÝ\²µ±l‘Ð.¢‚9h¿ÛðÃ# ŽI'meÌÇãÓá”ÁùÌšÜÌk҉œ67('œçƒä?ӝLu> 
     447úÞ“»Šb78ì=•‚-·ÔëuÓ¶<J2%6;þ¢’¿n[:¢Èkxÿ óš$7uœ5YNK¶ƒŠˆí?Î<'}QR_ƓãglB²ZŽBXÀib±ÔŒº¹؄ÊX"(Œí"-Xø+V›êaÓ+Y·‚SHGJýQ·øß±Œ‰Ê~¿ŸíaBiïC9td@aXàõ        r$Cëþ€‹/\&Q©º»‚œÇ•õ2 ¯cë4á¬A³LÑã&øœämœ 
     448÷hqž|ÈÏ"Aä 
     449ѵµ1ÅÝ 
     450Oðü 
     451EDó‘ 
     452þ1*›À8©Ò`nHß{Æ kÀ˜¥²¿, 
     453ÊbÛt‘žÌÁ“ŒÁˆc$œOðîdù}ðþÝEž1  
     454ç’Åb’'ô¿ëOyqU‘nãC9²vœlB}ž–R yˆî 
     455©      ùMÆOy³Wa§³‚ƒh^ÊÁ[åÅBã}¹£4 
     456ÔҖ_&âl4žmîƁ6Ȳ¶t/Îx 
     457_»e•€zŒ-än 
     458±Ñ=î€%9iÚJ9ŒL˜ 
     459€bä;‰{ýøj9ƒä   eÖç¢þ¶MîT®ý9Ô"£®'¯‹Li—ÿ)ü–Z™;‡T€b€?P–ppÎg­`UÊeçŽÏ¶b¯•Íóy+™/·â~òooÅsÀ;+Ž?ʉŒò 
     460ņSVü™šŸeųšƒGšç­øOà·³âY€ÁŠ#€ÿ9iŧQ~3+¶9é;kÅ"ƒXß 
     461›D²æ!ncØÂ"štŠÝü9ì<ež3€ƒTŒž¥3iø!"ŸVæ;NÑlهèp+mëèk'²JŸG Ø ì+ÏÎ1ð 
     462O°9Xœ¶çR1ŸylÈË9à+™à4šò@W#O¶ªàÐUFÞ¡Sƒ-S 
     463g0~lÀA±ü°¬ŸËgHx^*Wf  £_9‹Jéʧ^þ 
     464’  èŒš'Id˜€UfÝòøÅÜÜüyK-¶ÕˆW5á°4U?Pµ€ìÁ« À|U~œÊ 
     465U™×˜EDa\žK€á¿ýÊê*Kjž·){ŸXÆǵD’'Ñ=åbŽõV堏qbçrÅ7 !+UU˜y 
     466Df‚ˆMDDw¢Õv¢ê„•ÝHMƒ_a†qL™1ЉDd}ŠŒå 
     467»?•|‚1摟ð1–Ž€†Ç¯¢   Š2Lº 
     468Bùë 
     469§¡Ê\ 
     470ü”R žA)R1J7®5¹Ô›Aٍ³Ë 
     471íqÔæÕ6'·G„ž¿=ý3ÛA9œ=”oŸ=³(ÃöD(íÉí9rõ¡ÆŸ™miâ+ršöª(ü|™ï›C¿Cžö¡ G$¯µžu_ 
     472­M 
     473Ñ0›Ýˆ åé_%‡Ôð•vœëj>zý œwt/Ç 
     474m(†/YdÕÃ;nª»_VéU®aC>še®œH$|‹ÞGÃ]75—ÖÍT‚t¥Ž‹ˆ‰jÔAD2_ÏyQ@ѝÏs.22ºåŸ'žcòÐK8>DEÕ§Q‰qˆ˜»ç&3þã‚J•Î."¥ÆÄ9ù˜¯%.÷ÊØbž8‘™ nÂÎîz¢šî^Ÿ¥ÖQMµé™ßóã۞Ҡ    –l¿5Ñh`w—ŽÅŠ“Ÿ3^w§àþdµ‹J£Ñ   ËCÁïñÏ{ÞƁ 
     475~'þ•/Àž“?}<ù³•1pÒÒ~úw+ˆ\ª2˜óXh2Õd:SUéÑLÃUÿ$ 
     476œ6 
    477477endstream 
    478478endobj 
    479 146 0 obj << 
     479147 0 obj << 
    480480/Type /Page 
    481 /Contents 147 0 R 
    482 /Resources 145 0 R 
     481/Contents 148 0 R 
     482/Resources 146 0 R 
    483483/MediaBox [0 0 595.276 841.89] 
    484 /Parent 120 0 R 
    485 /Annots [ 153 0 R 154 0 R 156 0 R 157 0 R ] 
    486 >> endobj 
    487 153 0 obj << 
    488 /Type /Annot 
    489 /Border[0 0 0]/H/I/C[0 1 1] 
    490 /Rect [372.194 691.505 397.859 703.185] 
    491 /Subtype/Link/A<</Type/Action/S/URI/URI(http://www.scipy.org)>> 
     484/Parent 121 0 R 
     485/Annots [ 154 0 R 155 0 R 156 0 R 157 0 R ] 
    492486>> endobj 
    493487154 0 obj << 
    494488/Type /Annot 
    495489/Border[0 0 0]/H/I/C[0 1 1] 
    496 /Rect [316.488 596.661 338.487 608.34] 
    497 /Subtype/Link/A<</Type/Action/S/URI/URI(http://www.scipy.org/Installing_SciPy)>> 
     490/Rect [379.545 646.426 411.002 659.224] 
     491/Subtype/Link/A<</Type/Action/S/URI/URI(https://projects.scipy.org/scipy/scikits/)>> 
     492>> endobj 
     493155 0 obj << 
     494/Type /Annot 
     495/Border[0 0 0]/H/I/C[0 1 1] 
     496/Rect [132.869 619.328 317.444 632.126] 
     497/Subtype/Link/A<</Type/Action/S/URI/URI(http://svn.scipy.org/svn/scikits/trunk)>> 
    498498>> endobj 
    499499156 0 obj << 
    500500/Type /Annot 
    501501/Border[0 0 0]/H/I/C[1 0 0] 
    502 /Rect [88.295 548.094 164.658 562.826] 
     502/Rect [88.295 517.203 164.658 531.934] 
    503503/Subtype /Link 
    504504/A << /S /GoTo /D (id2) >> 
    …… 
    507507/Type /Annot 
    508508/Border[0 0 0]/H/I/C[1 0 0] 
    509 /Rect [88.295 481.478 319.705 494.086] 
     509/Rect [88.295 450.613 319.705 463.222] 
    510510/Subtype /Link 
    511511/A << /S /GoTo /D (id3) >> 
    512512>> endobj 
    513 148 0 obj << 
    514 /D [146 0 R /XYZ 89.291 778.025 null] 
    515 >> endobj 
    516 155 0 obj << 
    517 /D [146 0 R /XYZ 337.491 613.322 null] 
     513149 0 obj << 
     514/D [147 0 R /XYZ 89.291 778.025 null] 
    518515>> endobj 
    51951610 0 obj << 
    520 /D [146 0 R /XYZ 89.291 569.806 null] 
    521 >> endobj 
    522 122 0 obj << 
    523 /D [146 0 R /XYZ 89.291 569.806 null] 
     517/D [147 0 R /XYZ 89.291 538.915 null] 
     518>> endobj 
     519123 0 obj << 
     520/D [147 0 R /XYZ 89.291 538.915 null] 
     521>> endobj 
     522125 0 obj << 
     523/D [147 0 R /XYZ 89.291 507.41 null] 
     524>> endobj 
     52514 0 obj << 
     526/D [147 0 R /XYZ 89.291 467.877 null] 
    524527>> endobj 
    525528124 0 obj << 
    526 /D [146 0 R /XYZ 89.291 538.301 null] 
    527 >> endobj 
    528 14 0 obj << 
    529 /D [146 0 R /XYZ 89.291 498.741 null] 
    530 >> endobj 
    531 123 0 obj << 
    532 /D [146 0 R /XYZ 89.291 498.741 null] 
    533 >> endobj 
    534 126 0 obj << 
    535 /D [146 0 R /XYZ 89.291 475.438 null] 
    536 >> endobj 
    537 145 0 obj << 
    538 /Font << /F53 149 0 R /F57 152 0 R /F15 84 0 R /F44 75 0 R /F63 160 0 R /F64 163 0 R >> 
     529/D [147 0 R /XYZ 89.291 467.877 null] 
     530>> endobj 
     531127 0 obj << 
     532/D [147 0 R /XYZ 89.291 444.573 null] 
     533>> endobj 
     534146 0 obj << 
     535/Font << /F53 150 0 R /F57 153 0 R /F15 84 0 R /F44 75 0 R /F63 160 0 R /F64 163 0 R >> 
    539536/ProcSet [ /PDF /Text ] 
    540537>> endobj 
    541538167 0 obj << 
    542 /Length 2788       
     539/Length 2910       
    543540/Filter /FlateDecode 
    544541>> 
    545542stream 
    546 xÚÍ\mo 
    547 ·þî_q@QÔ.|̒»Üù’¢6Z MÐꛊu–T[>U’ãæßgžœÙ 
    548 Ÿ 
    549 ¥œ ŒµN·Ãy†3;/ 
    550 r¥w 
    551 üÓ»qRfÒ»aUcìîÝͳfw     wÞ<ÓHÑé^õS Ÿ37÷twßjeÆvŠQ 
    552 ÜT†žOw—»è›²ñˆ¿º^ 
    553 qã¿9{öÕëŸÛéFMÍŽ;{¿‰4goŸÿáÅ^ÛA??ŒÐÏàz—Z®œ5íó¿ÁÇOþ££ž‚ë®{ÿ•p>ÿꙌ )þן‘êŸÏ<= 
    554 ’Üà׌ìþÖ=2;Âõ®p]ø[?­ >ÀÇ?¯wîpÀ—UVBEŠ‹?žýýÙ_ÏVMZµÆì˜SC5F5-|Э²ÝŽÝPm«ôÔʆBnšw8­š‹›yzÑDt7ª&À@’‰M*I›‘DOªïºY’/    š6 
    555 d[ˆf•tNô߬’ிg RIb'Êש(ƒ‡6%™8ÒTà Øwœž$pfRmÓoŸ9Ž¯‰¢8QT"Jk!òyæHS‹898òÝ;tÒ_àú¡± 
    556 üx›H2Àøö¥àøŠ”Š“²ÁÀe‘&xvF++i*°'û2†3M«z=n֍—E‰Qž:žl"–éU7v¢ˆŠ 
    557 qÊk¡ 
    558 Ô€·?4Ÿ"J„µÐfµÐ7jeG!š 
    559 tÄÉAÿè 
    560 C;T÷(kÕM˜,ìLp3ÇíRmzÕL§DЕA‘HŽ L”rá(‚2žrÝ:ó%‚Ê¢„(r•fŸDP.âT ¹(:ÂÃeÛíŠÁñI#'i‡Þ£Ó( 
    561 ®ÖT„4؈S!Š‚ÑZ»Y4^%Fq¢Ž‰( jµ8s¢©ÀEœ(f`Ùü¶D·*a      ¢¢TŠ“*1¿é ߎrd š 
    562 \Ä)o~ šêO0?Ž¯ˆ¡8QÊ)Cšù’2DžˆSÕü#hªÛžh|EªÅIµOEGé³æ¢HžOM€͉””uf㍲Y<I,ä“} 
    563 !}(«ÍfKÐxY’%k‰òž1Óɖ >‘"Žl€l¡"˜:+š‚hjp!'ò°°2fTœí–Už£ò=†BeÕCÝn;»œ²b ʕ‰•¥XYÉpXYqžbeµyæTYUD       QÄÊJœ9UV2\Äé”Êj³bš²’%P²)U°Žê'Y1HSƒ 9ec•]›gNŝ,J„âgîkJ[ª¬$ 
    564 ,••szleµUKe%K¡äËK¬‡D%Pe%ÃEœÄÊjóÌ©²’E‰PüÒ¢PY‰3§ÊJ†‹8=¶²Ú¬ª¬d©"”|ƒµŽšªš*p!§B%šš¶?ø4^%Fñ~! 
    565 Pҗ4°”2lÄ)W>pdìo÷œU 
    566 Øik7Ÿ/vó‰Æwó5Ôu¿÷+¶’Û›r 
    567 Í”•†)­Š¡;y{€·ZÓ(+ip{ĹùkÜ8ºÃ 
    568 ©‡uo‰vžh#é®WŽŽŽ:à®Ôî±üâIÞãíãŒå2uËöŽÎq'†@L†W‰ 
    569 uÃæ˜ê±í•±#u7«±u›<5" 
    570 Sã78S¯/Q 
    571 lkín݋s”ÿÅ 
    572 *§YÚÿûˆ\>!§û9"$I펠  ŸI 
    573 ž&af÷ª7 
    574 Ä"¿Jžœ-U*P‘ðÉ 
    575 ÑRŒ® „╉“®:ˆ›íˆ’nÝyš 
    576 \%!<; 
    577 î 
    578 ÌþÛÒ¢c 
    579 ¶Ï 
    580 ÇWĉPò+ 
    581  
    582 òeÓɳGš 
    583 \ÄÉÁ 
    584 N 
    585 >•ŸñÉŒ_W_Öòá)^†ÛÙBF±PÞì¹l|Ñs‰æ÷›Q’ÒcT] šŽÖŸ”†Ižš?ìØÁŠ\>A4,ðý‹e€cxZ —ÃÇõ<=vLáÜÿðî7ŒFî÷È2<‘Ë€ 
    586 •›YŠFò4éô3QÑ 
    587 xˆ©GE"Ü 
    588 ƒrTäâ£"¥e8ŒŠ 
    589 Î›eñû‹ÕÑ?„!€’Þ¡ŽµºUU˶«({Œ’U•i2’¢¬àñǐ@a Œü³ŒÁ7Ù¯,#RqÿTZðf€|šHŽ(–EŠÐœHì ÒeZ€€ž§ñtñOEŒ-¯±¬ª;шŠqrpR 
    590 ƒŽÝØíÒØør FˆÏÂóc÷,øþ®\íÍ€]±0þaœE¡û|]”s|tX̧üáȚŽT6ÜI#@=&áä"‹’Äàn˜`X!9‘æIRMçb>ž‚m:!æ»e»éÂ'Ç| 
    591 ¿g 2A8BÉׂ 
    592 øE(Iò=I 
    593 ,àã° 
    594 ó8ý&E èŠßî<ëø²ó  
    595 sžïÒӕOp ÌáΏléƲãOÿ¯e¢Ê3óŽ 
    596 Æy{Ž˜ê“2©‡Ô¥Ó׶1jŽë  
    597 ³ŽPH2Õfgjß/ç^—˜uµh( Šªò²Tv:0g¶ÿ­Ž¥ÅáÃÚü+ـLö)eÏ\äŠ5=’êšk!B 
    598 vÇ–ÚĂ²ÇÓGk6r'Žh˜Q¢óʹÉVÝŒÃg”>߯†¹^Ü%ò» 
    599 LÞ 1€,n–ꗐs<ßç[`Çe‰ÃhPÆžcŠË8ŒQ 
    600 N7Îŧ•ÛQD̳Ù- 
    601 ÍÅüÐCÈŸ‡ì¢Ž‡(*)XoðžYïv‰^œ}žvp±Œ„ØR#Ž­êÜyŠSÆ4Í$i˜2 
    602 ÐÃ|DŸ^ð- 
    603 ¥ËËÐc 
    604 ¢þO°Î%K 
    605 ‡¬K\á®ðmèŏkà&¬¬¡âgZL-«ïÑíXÈEÓ      –‚¥±–[@HÂìô—(J<$OãŒóy‰^ü¥Œ™äÛå Ïàsè@™gá‘-× 
    606 ÏÛyJ|f„º®Ý­ZÊžÌ4·ANv£•í+–@f 
    607 ÛåZBé€Õ\ 
    608 ^¯n€yv 
    609 áÑ .ó¯_Çq 
    610 k€`‡ã†VdöãjöŒÞó%n×Xe‡aÇt“yYÆÓ€ 
    611 Ì¬Mºf„$³.ÏϗÎX¡/E6÷¥ƒr_Š‹UìKqQŠ})ûR 
    612 .¿BÃ>úæÙS·^'B»õâì©[/ÃEœh‘Vx›Ò[•²Œq!J£ÐɖÕÇ mAIAKëO†Ž8Uº6í#u·9à±ñŀG4,à}f:+ޒQÓWØŠN5gùLÒÃæž&7^Ü¢¯íœžßÇhî(ÔÀDóÔLJïÙøTŽ£Ðóð.HÚôÀ‹ˆsrp·y‚~÷9ۆx;o?èÌù1|çj«š–7»dÉ#ÿFO៚(:š'ÃEœ¢ƒCVu™·*gée‹ÒÆ( 
    613 “ú°”ƒrGRÑÔàBN«r 
    614 -ó`ö[Zå²8ŠØ*ÏÌþi"Q;])B˺µ5µú7É󩉢嵄'ÓÄg„ΞÉp§Zj²­õö®_NMHó€Ôt`ÎkLO÷8â:lc 
    615 çîC°ð  ×l™õ`ÃÅ._‘Qüyéð»Ã|dnþøÇRεZ¬ÂÕåÒ$T«àfõ 
    616 "ÜŒCÀ鋉SÚ!à’”v*`!UŠ†JJb˜f»rüðŠž!F~aBIŽ¬·<Ð>ð–óﲐ”V™”ô^î=    ŽÅµdÍ¥L¡cN< }õžÿ•t¯uºtŸ1©Æâ8žZz}fpžÚ¶f} ü±5׍‹ÿ 
    617 †ûê‡F»÷'/—†É«BÔºWCž­•L—B 
    618 £gLÁ@<é•aö„¹ 
    619 0ÁòA@ý›\Q 
     543xÚÍ\[o 
     544·~÷¯8@T.|˜åm/ò’¢6Z iÐèÍ       
     545E’%ÕÖ¥’l'ÿŸÃ%‡;ŒÍJ{‚"0ÖÞ³;œo8Ç\Ë]änœ„šänFÑ)»;œ~Ñí.àÍÛ2Pً~Òp_y¹Ç·{- 
     546õL#:x),4qw÷»ìÉ¿Hû€¿L/£\ûo_|ýŠ7;ى©›vÇæøìÝџ^
     547äÑþåÞ*}ôù¥<:ë 
     548®«p×)\çp=x2$9_~~œßÊ£[ÿÈÿ 
     549× 
     550\¡129ó$ õ/ ¯{‚úÁ?Rp+ü훹Ÿž1/«›ðÛ ôʓácá 
     551}yùóñ?^üíx1ÑhEßõ;¢ÄÂPzê 
     552–v·—ZX3m7”ÖBNš7T ¡†ºZ¬uë‰ÊxŸªÐýü(•3ÎåÂû}à}¬üˆ™B[#äiìÖFß«Â`Ó :=íˆ2KÏê0æxžÁä$zcxƒj°óà§ó(Ækîïßçþ% 
     553"æ= 
     554xˆœ¯©’×ÙLr 
     555 
     556?²_ý«‡À얌‚³Â'?Àí_–7 
     557sß,g¥¡€ÐJíˆ+†R`Lé 
     558e·Û©ëE7­Ø)ÐP;†^} ªžž{—õCšQhˆ €èҔ’èR’~R0*í,I€Td‹D³JŒSþ5«$yBU‚í÷„A)IŽâDùŠeã QŠŽš8æ 
     559§îûNMBC$ÞÜóÐ~M”ʼn" 
     560QŽ 
     561)]ñ=4+p'‡®‹ÓÞopýÔÙþyWH2@{}€RBû)3'eâ–*D"9Z^9f6ãä`‹x¯:-z9nÖ¶çEÉQš \¶KõŒ†ÕÒ¬@gœêZЃ˜äöq€íWDÉPštU }'†‘w€YÎ89蟜cH‡ê†²f’qRw×sàn 
     562PḨ©? „. ˜ˆøJDi‡PC(k‡Ð­=!”%EáC(×óBYžŒÓj­ 
     563ÑF—ÕÛÚ¯Hš¡8IMpY†QðµnEAf6ãÔ£`4m7kÛó¢ä(N]ˆ"Á V²=GšžŒ6¿kFÑ­JˆQ”•*CqRæW& 
     564ÍG€YË8ÕÍoASýæíWDÉPœ(í9ƒëyœ3XžŒÓªùGДÙ>1`û©2'ÕŸ 
     565¥¯šÿ™"y>k"¥hN€"¯SoäÍâIVÀR>ÕáӇ°Rm¶¶ç%ÉQª–Ð0+5 
     566l       ä³"R†V 
     5672‚ÉXÖH³—rBI3+¥FÑ[“dVŸÆÐʬz+ ñí™Õ€ɬŸYQڙ ‡™kgV[{ 
     5683+^” 
     569ϬžžÇ̊ 
     570Ë8 
     571”YmUL̬XI3”ê”*GXHõ¯˜@³—rªÆ°˜vmíyLîXQ2ßsŸSÚffÅh`ɬ8؜ӓ3«ÊX2+Vª ¥ž^b>Ä)!fV,\ƉϬ¶ö<fV¬(Š_ZŽ2+®ç1³bá2NOά¶*!fV¬TJ=Á\‡SB̚xž”S#“MMÛ>¶çEÉQüÀoD8é3XÒ6ãTK(2žµ«†›Ëù€}³œ4Ÿœï63ÿèWn%·ëë<š(« SRLƒ9x«W†ÏÄ+4ЄýÜy­lÙ]¿\vžp#ÉmDŸN÷$ïÉðcØdùmÙ Ä 
     572¥kÿèŽìi„­˜s²‘Œ2Œ.ôšì¬ÒÇRºÊ 
     573ŸØK°™åÕèIˆ¿ 
     574 
     575õêb5XßHu”ÿ 
     576;TN±ŸÉïy؄{ êåœÃËÕØ£è•)Á“Ý«ì]YHG†NÎý»˜ÍÔÈN‘ðÙYf®„A;s¥â”KAS‰(åƝ§Y3Z›Àœ 
     577ÞW@ZXqŒÃöއö+âd(õeB“egøÞšžŒ“ƒK=8 
     578“ŸâÁ‘žtø²äׁâUº—ÍL'vè_möZÒŸé¶HóǝNŠŒc&QM™èOBB' 
     579zF[gùS+HCÂޏ$üߊGêá¯v@C 
     5809£pâÿñî7ß^îe~„€œ2€–€g•ÉӔݯEEX(ØÎ<!*ÂíQqaÀDE"N;*QÚQ‘ 
     581ÚHàŒY¢ßŸ-Žþ! 
     5828坷6V·ª*nº²²ç(UU)­„âeÅCğî 
     5830jðÍ2&Oª«]^Æ@Eeüskµ[‘òy"ኘ)Cs"¥gЊôô<‡‹ø¬ˆ—¡Õ56€U¥a 
     584iVà2NŽ›ÇÔ$ŠN9‘öí@ 
     585hH >N=àûïpZ3OŒÉiŽKÆ?,¯0tŸ,˂ö 
     586ŸóS~Dž%GLî’I#Aœ-ÂÉY¥ˆÁf˜Ä8%V(Ž£y’RÓµ˜/a¬u=óݚ]™Høì˜Úï    ƒJÎPê¹`/SIŠïIÖÀ>ë)Ãé÷H+ÆtÃfç!í›Îƒ4ÄyþY 
     587­|†UNv~$ 72;þòÿZ$ŠjPP3ïÜa\æC‹ªŸH“zXžiãÀ$Ó 0—ZŸäƒ4Ä<L&Itÿ‰œš]Î>Ç uU”PcZyÑÊ; 
     588˜³Û’BGíÀ5mvS²'>rIJ 
     589Ezd4„šÁîˆÂJ£XPöxxÙÈô0¿u|ÙiˆQ²ÓʵΟ/º9 
     590ƒ=?,†¹Šþ’9^ 
     591Šî†!ŠD?+õ‹ÈÏ8³Ÿ 
     592Â/<F‚§Œ;¢žŠÇ(ÑÉápã˜Qô=oOBL³Ù) 
     593 œåü‚ uë}Р
     594© >ÛÝÅàEÙ§KG‘›@)˜ûÝ¢žÒZ £í¡‡Î¬yxb€Jå?›¹"KØ,8Žª+¯R_yÌNö?>Ã2dÞšÀqvƒïRÿ}Zá¶ÈÅ`u 
     595Éþ¢ÃÒJ££ÛšR0 MÛ7L7‚¿ñÕ€!†úk  
     596‹¡ø6Œùýc&ù.~yá|Jœ§2žÔjUñz¢]w§ÂaFÈéôŽè©â1“V 
     597 
     598³ôd„éùRÒ[žøÝÅ+ 
     599TIž*œZ 
     600!̱ÍOÃæŸßT?ŽRéÞÆ5ɪÐÝ늯緊³ÂÎèŠò™Œ§)XY˜èžÈem~Ëb¢6Ø\”" ÚE)*V³(EEi¥xžP”¢põåY(¢oî=–êyq2¶TÏö 
     601Kõ<\Æ   WhÁص>¶Øª”ø±+eŽ‚gZoÔ9źqZ)Ùh«Å(·ç€};àð~fºmf»-£–_¯AÄè`Š¥=)™{šRÜZx7T!«ü!GsG?!ÿEšçŸÐ|Oڗrd‚‡¯ŠPAʊGš¯°h9'w—˜')vŸ=ˆwóÞCåVüÜj«šâG]ŒäŠÿ˜§q|U 
     602Ñãá2Nّ¡'+KÌ[• Ù¬Ž9JãŒþ$4€;œrf 
     603.åŽ(§Q/Oz¿¥N΋“¡°uòJïŸ'ÖÒy‘2Žª[«a“–¿ƒHžÏšH)Z]KáL;Fðt—qZ›š`&äö‚(iߞšͳŠŠsR匠
     604ÓÓChq•Ö0oçÚC²pÊWd‘Ÿ`Óå.]’á,ø9VèÛa>,7ß~Õ̉VÛYx®ºÚ4 K%ُëÛHžy{€0`Š/"Nk{€JÒÚ 
     605XKø0Seiš"%†€©¶+Ç7_7Åš/LpmëÆÎc¡Û~ò§U"%~’;TπcrÍXsIÓ9蜠
     606H_¿4q!ÝKi@—ó 
     607UtÖçÕ_.c¥O 
     608ÎSµVˀð'Ö\1.ÿ0Ü£Ÿ:韜Œˆ“׍0(%øb¶±œwC 
     609¡{›tAA<É 
     610aõl¹ 
     6110ÁòAýG<»ö 
    620612endstream 
    621613endobj 
    …… 
    625617/Resources 165 0 R 
    626618/MediaBox [0 0 595.276 841.89] 
    627 /Parent 120 0 R 
     619/Parent 121 0 R 
    628620>> endobj 
    629621168 0 obj << 
    …… 
    635627>> endobj 
    636628171 0 obj << 
    637 /Length 1558       
     629/Length 1574       
    638630/Filter /FlateDecode 
    639631>> 
    640632stream 
    641 xÚ­WKsÛ6ŸëW°7j&D|¥·LÝL2ÓC]:iŽDË#Òyüûî $HQ²“v<–€ÅbñíâÛ]H1üé (•)uç 
    642 ŠMl›U 
    643 ìaåÍJ‹†Õ™ÊÊÆ ‹‘[’Äš"ÍQ)ø 
    644 Ä Ž©!- «ŸWYLâÀŠ±Ÿ|Ûèà·võ~Œ žÈÚTé¬@ÅכÕËßuèX•qln`+Xc”Nl°Ù} 7·k 
    645 ÖGüX®$¬Œñ5 «î>·Õ=|âÿw\2a¿6Eøumò°Eåq 
    646 ßA«c= [ 
    647 nÉnÕ{'ñאüÁFñ€îF­ 
    648 GÛú/Õž†bÁ˲Ø<Ð*@xía<çï8q¶çéWßRŽþŽyD¡H'*µ%Š}$µ(Iã°©      Ì ž‘§„g‹Göâ/,‘ÚŽÇÛõŸ¬ÓŒcz7úÔ$ÍZ×4Œ÷šòETQsÏò[ÖÑáÃ:2{bô™,WMÝ»[$ ÚùŒcMMaÅaë<±%œ†ÑÏ¡^O¬·7šx”-2àÑÝø·e.úH–‚úŸ TÈƘᖆ™,%' 
    649 œE‰ îZ YîEÕ9KPŽln)ÊhKNºçB‹k.’Œ÷4HK!eÂ1«:Œ˜Ò†lï–óB‚d)$ü߈¯m3ÆÌÅœîoØ 'XžlmeO×óò™cî„!€¥ ¹”†Er¬Ÿ‘“îYX‰z%ÎŽGùÊŀªž‚|{Êü£š¡COÚŽ™r¬@p9Þµ<9uŠœ@-oŸX<:rì8ª$” ¹w{–ŽÑŸ“L†s~YSBižM;8[B@z[ě­(ŽÑBh:I„È&οIr€Ç5{ –Ü"Š¶nïµË:”î 
    650 õiž E=v¢ª»nÊ 
    651 /mHwEžÇ'ëŸ>žHÈÁÉÕÕfl‰&Sº ’Ôª€(¹©}ü;èïV±²I|]aç3ºššeVf÷«+¯Å‰©È³uÚâl«4µÃeïÊ$nrûÇgHQj‰ÕÊd     ßSJ–˜GŒÀ Ì 
    652 –Fœ@“-²|ñ!àzµ§CŽXh­J<ôf°’Ìfã5d£H Ô²ç=Uxºwy‰”^ün 
    653 «úÇcÍ¢‡jÄá±Â¥º¯ 
    654 [»9Rr`É­úêœ_q©€Þ—ýuN/H”à_ř&¿Þb—(¢ÀShŒ4á 
    655 ”’H’›¢œ”Z«X>fېYžÚ,ŠoñÊ-ÍwŽÊbÒ{/­+b¿¡¶2ÇL¹PïNàriåñäå„ÇŸº[(>º$·@e#gÖÍ+¹tÄGTžŒo_ 
    656 ™JÓ¿ÖEBXqÛŒkä'oŽœË~žÎh¡Ú€6üõ¶²wt²“4–^w¿òÒP” !"œX)x¡®c PLÚ®M&Aý²•Þ†ßŒc¡°Ó²ë-ÖÌÏžrößåý-Ø1éqú($’ŽScá 
    657 ž+bbúþâ2䮣ŠSVÇN|h: iKRrÄré®wŽcÇz3 3à'oª 
    658 -oÆC—.›•š.é)mÿÒË¢›»¶ð" 
    659 wÇïÿå&}®Vš±Ð†°€)(Ç1ü: 
    660  :ÅÁqÌ$.”»q;U;h 
    661 §ÕŽU&MU{Hkøi¿ =D³"›ª\gòÑÇv"ñÁj«"O.£uJ 
    662 ÜƒD·¡§œŠ'Ÿ*¹§ÝYwŠLÙ¬|®;?{]B Šô       wDÉsçÎs¥7ú9l 
    663 [Ø °#§rñ|¿ózG,æÿÝ2êÂÂóÃü4œýgéètžÍGÔÿMȋ€–ž£W#Ÿ’ÝŒ’×âÕ7Ywºõ’§ÆÄ*Ϟô4ÿ  ®\¹è©Óò=} 
    664 [3@^˜KÈ#§ó„ ”Å‚+ÓPñSã·ÕAÄ­æ ™s8Wÿ2ûϓYtž 
    665 0m ÀOÜ¥«þRxùÆÙÀ–|+ˆ™€sÄ)€a®/!Ž2ø-QðÈL 
    666 ¯qö™®uŠòØ.?ÓMiU©±°bŽs\iœª9;ZšŸú/§8žh 
     633xÚ­WKsÛ6ŸëW°7i&D|µ·L]O2“C]:iŽDËnM1é€ù÷ÝHˆ¢d'íhD‹ÅâÛÅŸh" 
     634?¥²¥‰òŒPÚŠÑŠYèh+× # 
     635Îd*+Ï,Æ~5N«Š4GŠès€èRK\áV?/2MäÈ¥Ú_ŸnLôk»x·ˆÞÂ\ì\ªLV ã«õâåo&ŒV¥.£õ-ŒŽ²‘³V™ÄEëí‡åúne–õ+À•,«`|껇çŠz€'þ¿á’]ö+[,¿®lŸl‘àx\áãpu, ‡aËà
     636É­úà$~]å- œöž¹* 
     637mêŸy©Æ5$ ^–°—Åæ­$×îÇsþÔ©Æَ§_CIñêãúM‹ 
     638b“šÔ•d˜ö‘Øâ$ÕËŠ&0/xFšžÝ 
     639 
     640ًŸ°Dl[ 
     641oZäû²J3¶éýšS sÝxÐ0Þ!ËaEΠ
     642Óï˜Ç,?­b›±ö@FIrÕÔœ¿EbŒÎ[ÖÔdV 
     643¶^W õlê͑ôö–L£Œ²Axô7ÄŸÀm™·þ’©  
     644ÞŒ ù²!ÏðKƒ‰l–’Á¢XwÍ8 Iî 
     645Õ+KP,nÎÊ(K<œxϙ× 
     646œ%y祝/C–     Û¬êðbJ·dyw 
     647`t–ÒJ˜¥‚ÿ­èÚ6£ÍŒÝëNñ†õp‚ƒË6Nöt=/Ÿ9æ^<Žd4Ò°HŠÕ·rÒ+a¯D™vï]Ÿò6 VÜ        F>5ÈEþAØÓ¡&í8ÙrÌ@p)Þµ<9UŠµ@-oŸ$X<xçزU‰(r#êöL 
     648­}/‘ çüŽ¢€2<;vp¶˜€ø*–ˆ7[‘iãÓt±³ì8ÿG”wź³ǒ[DÒÆïœñQ‡Ô­w}š'C’AÆÀ;‘Õ_7ÅN6Ä;ãEž'tÖ]œ÷6“ƒ’‹«õXmŠL%©SIQrQûðQG[šoZ¹$‹Ÿ.°òYSDÍØ2'³‡ÅûEPâDT 
     649È:-q.Ñ*MÝp Z9ž2±›Ü~ÅöB”ŠA⌲YÂ÷”’€9Ï#@#L–Bœ@‘-²|¶ðµ:à!Eœ‹ŒQ% 
     650z;ÈJæ2‚ñ 
     651¢Q$8ä²ç=exºy‰€^|7Þ«úÇCÍ€OÕ*ÑËC 
     652Ku_:–v{ àÀ”[õÕ9œt©Àœ/ë5òœ^0Á_é̐^¯±Jè¢à§Pxi 
     653 
     654€¢HÛ¢<J)ŽV1}Œ¶!²pµ™ 
     655+Þ€[šoœ+‹È _ YWäý–ÊÊ3ÅBœ=Ë©•ÇGŒ÷ÕÝL‚Ñ9—@«ƒÉÀAø§C~£¥«!>iúǪH!N&µIÇ    ©fŽ€ž 
     656rLê–!ßFöNáUüº 
     657Íõt¿ðæ!ígìš8Is—Ÿxú²KŽŠíp07”À> 
     658Œ²š¢á›wÌ€sZö%ÄÉ;®Œ|Ööp ÖIjIÅud§÷€1]MŠˆ8îº8ùøë«%&evl›pB3ÙÍ8¢’ª@–K÷Õ·£ 
     659[æ›H˜?é€f¶\‡Î]63Q6Žð‚ÿ\?ÑMU›éC'æÝr×?_šÏeh ™Š&2IXÃ7Q9IG‡]4¡ü>“äÆí”ã `œæ8f9ŠcÊÕ@ÆÀ|ˆ&©5U¹Éä³Ñc{D  Á“«"O.£õLÜœX·¡†ÞPc¯D 
     660Jîi{V"S.+Ÿ«ÎwØޔ`š"}B 
     661a 
     662Ô¹TiE~ 
     663ÛB{á 
     664wvìY.žoáëކ§wäÅüߞqFS8h:ì{c°ÿ¬;zžgûcêÿvȋ€=Wˆž£¶‡³ÑßòæOovͱp%ší 
     665ÞmtÏ{áUò>±€MSèNÒ§,ÿ€ÛL«Œ°OXÀs 
     666xž—€Œž<–Ë 
     667Êì€OÞPRہÄ茋ç Ñü÷Ÿwqáy¶¹Œ 
     668ªŸ     !žKý=®n¡ ÖÙÀ=Wˆ|#ˆ9óN§œ¹¹„8Îà».>Ž.BŽ³-»1™Êµ›oÙméTi0/0btS\©NUYž(4=õ_„!º 
    667669endstream 
    668670endobj 
    …… 
    672674/Resources 169 0 R 
    673675/MediaBox [0 0 595.276 841.89] 
    674 /Parent 120 0 R 
     676/Parent 121 0 R 
    675677/Annots [ 173 0 R ] 
    676678>> endobj 
    …… 
    11201122/D [170 0 R /XYZ 89.291 325.594 null] 
    11211123>> endobj 
    1122 125 0 obj << 
     1124126 0 obj << 
    11231125/D [170 0 R /XYZ 89.291 325.594 null] 
    11241126>> endobj 
    1125 128 0 obj << 
     1127129 0 obj << 
    11261128/D [170 0 R /XYZ 89.291 302.291 null] 
    11271129>> endobj 
    …… 
    11921194/Resources 175 0 R 
    11931195/MediaBox [0 0 595.276 841.89] 
    1194 /Parent 120 0 R 
     1196/Parent 121 0 R 
    11951197>> endobj 
    11961198178 0 obj << 
    …… 
    12021204>> endobj 
    12031205181 0 obj << 
    1204 /Length 2668       
     1206/Length 2679       
    12051207/Filter /FlateDecode 
    12061208>> 
    12071209stream 
    1208 xÚÝ[[o»~ϯЇÊhÄ./»Ü=Ešà4HÑö¡ÇomQÈÖZ6bI†ÖŽ“óë;ÃîrwIÊVrZŽlí’Ùo® 
    1209 RŽ\ðO.êFšF.¬­E¡ÊÅõîM±ØÂ̇7’)Œ¬DÕhxŽL®Ê¢Mm«€Ã»Ë7¿ý£,²MÑ,.oz)ŠÖšÅåæïËrù÷³RŠ^^ßÃóºëð­Ynàå@kœž§çGxŒ 
    1210 Ÿ–^ñq}„_âñ|¡ìò€ŸhàÙ 
    1211 éÙŽŽz?wü¹ýfñý‘‘¬÷Äþú°{€á'7Õ `èpú 
    1212 ¬®‘E»¿PµgÔ!éú‘¥uþášuÜâËg€‡UÿŒüØs%µ(MãìÈш 
    1213 å 
    1214 /ÔòŠÑÆbZ8ÜÐ爃]îР֙¶EcŒ¥áõž šMåT}ªÊQ 
    1215 <œ}@ Zôà-¿ÞµŽêèçÈèøzØE`Ô)ƒ¡ø€Á~wÂ>e±ü‘BIB      'P\×:lL4ĞWnï%:{\3×GŠC 
    1216 $èZò ³ÅýØV4èµ88       ‚˜]Èïœý18iîŽMÉâo‰‰ãüe 
    1217 žBk”γ¶t‹ 
    1218 Nî;o^ÇQY‹P–Ü{G~Ðí(¢pv‡S_X‹'vgËs£(r2}átom|    Sdw,4ÁÎÂGp–BTˆ°G‰òoˆ¹WÙ»_PKP ,è­7y‰–5« µžûL«åၹ8⯟۠
    1219 MMóŠ#.Ÿ¶ ÉO”V0ô|ë£1ßÒàšêÊ%›ؖY»ž°E%|Ôí͏—}æÊZY-LŽ@3Áª§pÕ٘ 
    1220 4’.šÎUY ÙP@ýaóyœ¿FŽh 
    1221  
    1222 |yx mÐ e)ŒT1èŒBغÉC4FTJ@JÑ`Œ 5šQÖä»õ 
    1223 .