From zero to Matlab in six weeks

Transcription

1 From zero to Matlab n sx weeks Frederk J Smons Adam C. Maloof Prnceton Unversty

2 Explan the bascs I 1. help 2. lookfor 3. type 4. who, whos, whch 7. dary

3 Explan the bascs II 7. plot 8. xlabel, ylabel, ttle 11. hold on, hold off 13. sprntf 14. prnt 15. load, mread

4 Make a cheat sheet Name Functon Example mread Reads an mage fle x=mread('flename'); fullfle Constructs a vald path name ff=fullfle('drn','fname'); sze Queres the sze of a varable s=sze(x); plot Plots (x, y) values on a graph x=[1 2 3]; y=[ ]; plot(x,y,'o') xlabel Uses a quoted strng for an x-axs label xlabel('elevaton [m]') ylabel Uses a quoted strng for a y-axs label xlabel('roughness') hold on Keeps current axes for next tme you plot anythng x=[1 2 p]; y=[ ]; plot(x,y,'bo'); hold on; plot(10*x,3*y,'rs') lnspace Makes an array of N evenly spaced values x=lnspace(-3,3,100) between a and b reshape Changes the dmensons of an array x x=lnspace(-3,3,100); to a rows and b columns xr=reshape(x,20,5) hst Makes a hstogram (and plots t) x=lnspace(-3,3,10); hst(x) bar axs xy axs j

5 Explan the bascs III Addressng: rows, columns, dmensons, range 17. sze 18. transpose 19. colon 20. lnspace 21. <, >, ==,, &, Logc: logcal, character, strng, double

6 Practce on the command lne % Specfy the path - drectory strng to where the 'fname' s dro='/u/fjsmons/classes/frs-span/matlabdemos/matlab-lec01/'; fname='h1w-18_35-test2-small.jpg'; % Read n the mage usng a canned Matlab functon rgb=mread(fullfle(dro,fname)); % Convert to grey scale red=rgb(:,:,1); green=rgb(:,:,2); blue =rgb(:,:,3); % These values are unsgned 8-bt nteger (from 0-255) so % they requre specal attenton to convert to grey scales grae=unt8(round([double(red)+double(green)+double(blue)]/3)); % Make a pcture plot(grae(rand(sze(grae,1),1),:)); axs tght

7 count Wrte the frst scrpt 2 # # # red value green value blue value dro='/u/fjsmons/classes/frs-span/matlabdemos/matlab-lec01/'; fname='h1w-18_35-test2-small.jpg'; rgb=mread(fullfle(dro,fname)); red=rgb(:,:,1); green=rgb(:,:,2); blue =rgb(:,:,3); % Make hstograms of the colors and annotate ah(1)=subplot(231); h(1)=hstogram(red,0:10:ntmax('unt8')); xlabel('red value') ah(2)=subplot(232); h(2)=hstogram(green,0:10:ntmax('unt8')); xlabel('green value') ah(3)=subplot(233); h(3)=hstogram(blue,0:10:ntmax('unt8')); xlabel('blue value') % Annotate; set the x-axes to reason and y-axes to the same lmts ah(1).ylabel.strng='count'; h(1).facecolor='red'; h(2).facecolor='green'; h(3).facecolor='blue'; set(ah(:),'xlm',[0 ntmax('unt8')],'ylm',[0 1.1*max([h(:).Values])]) % Prnt the pcture for ncluson n a report prnt -dpdf madson02

8 Wrte the frst functon functon madson03(dro,fname,szo) 100 % MADISON03(dro,fname,szo) 150 % dro drectory200strng % fname mage flename strng 250 % szo sze dvsor rgb=mread(fullfle(dro,fname)); red=rgb(:,:,1); green=rgb(:,:,2); blue =rgb(:,:,3); grae=unt8(round([double(red)+double(green)+double(blue)]/3)); % Plot the szoth part of the mage n color subplot(411); mage(rgb(1:round(sze(rgb,1)/szo),:,:)) % And plot the same szoth fracton n gray scale subplot(412); mage(repmat(grae(1:round(sze(grae,1)/szo),:),[1 1 3])) % Prnt to a fle wth a flename that you learn from the functon name prnt('-dpdf',mflename)

9 Code hygene 1. If you type t twce, you need to use a varable 2. If you say t n the absolute, you need to reformulate to the relatve 3. Annotate all graphs completely, and gve them meanngful names 4. Annotate all code completely, to a rdculous degree Documentaton/Help/Date Input/Output Computaton/Algorthm Fgures/Embellshment Varable Output

10 Fve ways to do spectral analyss on tme seres Inspecton Play wth the plot, the axes, the grd lnes, the annotatons, etc. Correlaton Guess a perod, generate a synthetc, calculate how well they agree Inverson Guess a perod and conduct a formal regresson, calculate how well t fts Stackng Guess a stackng length and see how well the result fts a meanngful oscllaton Fourer analyss Do as the adults do! Inspect, nterpret, and understand the results

11 temp ( C) temp ( C) Shfts and Cycles I Inspecton /27 10/04 10/11 10/18 10/25 11/01 11/08 11/15 dates hours functon madson04 % Read the data and dentfy the varables w=webread(' tme=w.var1; temp=w.var5; % Convert and plot n a human-ntellgble format dates=tme/24/60/60+datenum(1970,1,1,0,0,0); subplot(311) plot(dates,temp); axs tght; grd on datetck('x',6); xlabel('dates'); ylabel(sprntf('temp (%sc)','\crc')) % Convert and plot n hours snce the frst sample whch s last hours=[dates-datenum(dates(end))]*24; subplot(312); plot(hours,temp) set(gca,'xtck',0:3*24:hours(1)); axs tght; grd on xlabel('hours'); ylabel(sprntf('temp (%sc)','\crc'))

12 temp ( C) Shfts and Cycles II Correlaton P of 24 h wth phase h has rmse 4.1 deg, corr hours functon madson05(p,ph) % MADISON05(P,ph) % P Tral perod (h) % ph Tral phase (h) % Read the data, dentfy the varables and plot exactly as t was done by MADISON04 w=webread(' tme=w.var1; temp=w.var5; dates=tme/24/60/60+datenum(1970,1,1,0,0,0); hours=[dates-datenum(dates(end))]*24; subplot(311); plot(hours,temp); set(gca,'xtck',0:3*24:hours(1)); axs tght; grd on xlabel('hours'); ylabel(sprntf('temp (%sc)','\crc')) % Make a 24-hour snusod to overlay on the data, calculate RESIDUAL and ROOT-MEAN-SQUARED ERROR A=std(temp); m=mean(temp); tpred=m+a*sn(2*p*[hours-ph]/p); resd=temp-tpred; rmser=sqrt(sum(resd.ˆ2)/length(resd)); hold on; p=plot(hours,tpred,'k'); hold off % Compute the CORRELATION COEFFICIENT between the synthetc and the data and fnsh the plot [r,pval]=corrcoef(temp,tpred); ttle(sprntf('p of %g h wth phase %g h has rmse %3.1f deg, corr %5.2f',P,ph,rmser,r(2)))

13 temp ( C) Shfts and Cycles III Inverson P of h has rmse 3.9 deg, varance redux 36%, corr hours functon madson06(p) % MADISON06(P) % P A vector of tral perods (h) % Read the data, dentfy the varables as n MADISON04 w=webread(' tme=w.var1; temp=w.var5; dates=tme/24/60/60+datenum(1970,1,1,0,0,0); hours=[dates-datenum(dates(end))]*24; subplot(311); plot(hours,temp); set(gca,'xtck',0:3*24:hours(1)); axs tght; grd on; xlabel('hours'); ylabel(sprntf('temp (%sc)','\crc')) % Do the INVERSION, PREDICTION, RESIDUAL, ROOT-MEAN-SQUARED ERROR, VARIANCE REDUCTION, CORRELATION argm=[1./p(:)*hours(:)']'; F=[ones(sze(temp)) cos(2*p*argm) sn(2*p*argm)]; A=pnv(F)*temp; tpred=f*a; resd=temp-tpred; [r,pval]=corrcoef(temp,tpred) rmser=sqrt(sum(resd.ˆ2)/length(resd)); vard=100*[1-var(resd)/var(temp)]; % Now fnsh the plot by plottng the predcton rght on top of the data hold on; p=plot(hours,tpred,'k'); hold off; ttle(sprntf(sprntf(... 'P of %s h has rmse %s, varance redux %s, corr %s',repmat('%g ',sze(p)),... '%3.1f deg','%g%%','%3.1f'),p,rmser,round(vard),r(2)))

14 temp ( C) Shfts and Cycles IV Stackng stackng over P of 24 h hours functon madson07(p) %%%%%%%%%%%%%%%%%%% Data loadng etc %%%%%%%%%%%%%%%%%%% % Remove duplcates snce we wll be nterpolatng later [hours,sort]=unque(hours); temp=temp(sort); % About how often dd we sample and how many blocks of length P can we dentfy? Interpolate snth=1/round(1/abs(medan(dff(hours)))); mblock=floor(max(hours)/p)*p; hours=[snth:snth:mblock]'; temp=nterp1(hours,temp,hours,'lnear'); % Now we are ready to 'stack', all of these are segments of exactly P n length hoursd=reshape(hours,p/snth,[]); tempd=reshape(temp,p/snth,[]); % Wth a lttle luckk these are ALL the same modulo P! f all(all(dff(mod(hoursd,p),[],2)<1000*eps)); hoursu=hoursd(:,1); end % Now comes the 'medan stack' and some dea of the varablty tempu=nanmedan(tempd,2); tempup=prctle(tempd,95,2); tempdn=prctle(tempd,05,2); %%%%%%%%%%%%%%%%%%% Plottng etc %%%%%%%%%%%%%%%%%%%%%%%%

15 varance # hour Shfts and Cycles V Fourer analyss h 12 h frequency (h -1 ) functon madson08 % Read the data, dentfy the varables as n MADISON04 w=webread(' tme=w.var1; temp=w.var5; dates=tme/24/60/60+datenum(1970,1,1,0,0,0); hours=[dates-datenum(dates(end))]*24; subplot(311); % Make a frequency axs ahead of tme nfft=length(hours); physl=hours(1); selekt=1:floor(nfft/2)+1; fax=(selekt-1)'/physl*(length(temp)-1)/nfft; % Compute a nave spectral estmate, the perodogram, and plot t S=abs(fft(hannng(length(temp)).*[temp-mean(temp)],nfft)).ˆ2; loglog(fax,s(selekt),'lnew',2); grd on xlabel('frequency (hˆ{-1})'); ylabel(sprntf('varance %s hour','\tmes')) % Annotate P=[24 12]; hold on ; plot(1./[p(:) P(:)]',repmat(ylm,length(P),1)','k'); hold off text(1/p(1),1e11,sprntf('% h',p(1))); text(1/p(2),1e11,sprntf('% h',p(2))) xlm([1e-2 10]); ylm([1e2 1e10])

16 Conclusons ˆ All Earth scentsts comng of age n ths century wll wrte computer code ˆ Computatonal analyss need not be scary f you take t slow! ˆ Learnng to mx pant won t turn us nto Rembrandts, but pantng mght! ˆ Masterng the Java whle-loop hasn t taught anyone the scentfc method ˆ Students wll want to program the computer when they have data of ther own ˆ Home-grown data can come from smartphones, local weather statons, etc ˆ Matlab s a sophstcated, yet low-threshold language wth stayng power ˆ Geoscences are a gentle frst step nto a lfe of the programmng scentst ˆ Codng enables students to take control and s a tool for dversty ˆ Home and student lcenses need not be costly, open-source clones exst