1. Scaling and Warping in Time Series Querying Dear Reader: This file contains larger, full color versions of the images in “Scaling and Warping in Time Series Querying”. In addition, there are some extra experiments which we could not fit into the paper.

2. 01020 304050607080 Euclidean DTW Uniform Scaling SWM  If we attempt simple Euclidean matching (after truncating the longer sequence) we get a large error because we are mapping part of the flight of one sequence to the takeoff drive in the other.  If we simply use DTW to match the entire sequences we get a large error because we are trying to explain part of the sequence in one attempt (the bounce from the mat) that simply does not exist in the other sequence.  If we attempt just uniform scaling, we get the best match when we stretch the shorter sequence by 112%. However the local alignment, particularly of the takeoff drive and up-flight is quite poor.  Finally, when we match the two sequences with SWM, we get an intuitive alignment between the two sequences. The global stretching (once again at 112%) allows DTW to align the small local differences. In this case, the fact that DTW needed to map a single point on time series onto 4 points in the other time series suggests an important local difference in one of these sequences. Inspection of the original videos suggest that the athlete misjudged his approach and attempted a clumsy correction just before his takeoff drive. Indexing video: There is increasing interest in indexing sports data, both from sports fans who may wish to find particular types of shots or moves, and from coaches who are interested in analyzing their athletes performance over time. As a concrete example, we consider the high jump. We can automatically collect the athlete’s center of mass information from video and convert a time series (It is possible to correct for the cameras pan and tilt). We found that when we issued queries to a database of high jumps, we only got intuitive answers when doing SWM. It is easy to see why if we look at two particular examples from the same athlete, and consider all possible matching options, as shown in the Figure on the left. From top to bottom: Video plays in presentation mode

3. 020406080100120140 C = candidate match Q = query 020406080100120 C Q (rescaled 1.54 ) 020406080100120 140 happy birth -day to you dear ----- C Q (rescaled 1.40) 140 Query by Humming: The need for both local and global alignment when working with music has been extensively demonstrated. For completeness we will briefly review it here. In the Figure on the left we demonstrate the problems with universally familiar piece of music, Happy Birthday to You. For clarity of illustration, the music was produced by the second author on a keyboard and converted into a pitch contour. Two performances of Happy Birthday to You aligned with different metrics. Both performances were performed in the same key, but are shifted in the Y-axis for visual clarity. (top) Because the query sequence was performed at a much faster tempo, direct application of DTW fails to produce an intuitive alignment. (center) Rescaling the shorter performance by a scaling factor of 1.54 seems to improve the alignment, but note for example that the higher pitched note produced on the third “birth..” of the candidate is forced to align with the lower note of the third “happy..” in the query. (bottom) Only the application of both uniform scaling and DTW produces the correct alignment. Click to Play sound files

5. Pruning Power The following slide shows how the pruning power of the proposed lower bounding measure varies as the lengths of data change on different datasets. –For a majority of datasets, the pruning power increased with the length of data, suggesting that the proposed algorithm is likely to perform well in real-life environment, in which long sequences of data are collected for a long period of time. –More than 60% of the datasets obtained a pruning power above 90%. All but two of the datasets exhibited a pruning power of over 60% at length 1024. Even at length 16, over 60% pruning power was achieved in three-fourths of the datasets.

7. Average Pruning Power The following slide shows the pruning power averaged over all datasets; 87% of data sequences of length 1024 and 65% of data sequences of length 16 did not require computation of the actual time warping distances.

9. Pruning Power – Raw Numbers Pruning Power vs. Length of Data Dimensi o n EEGERP Data Reality C h e c k ATTASBallbeam Buoy S e n s o r BurstBurstinChaoticCSTRDarwinEarthquakeEegEvaporator 16 0.32262 0 0.76457 5 0.84160 3 0.70150 5 0.88373 4 0.50457 5 0.876227 0.21210 2 0.912278 0.89890 5 0.70093 8 0.693372 0.50678 4 0.70606 1 32 0.36600 9 0.83511 5 0.87745 0 0.77104 1 0.93188 1 0.61660 0 0.912379 0.27317 0 0.928676 0.93935 5 0.75813 4 0.729497 0.68650 7 0.72535 4 64 0.37818 8 0.92345 8 0.92935 3 0.84481 1 0.96009 3 0.64818 6 0.959023 0.30566 5 0.934549 0.95785 6 0.77135 5 0.749861 0.72791 2 0.72851 3 128 0.40525 0 0.93547 3 0.97044 3 0.88962 5 0.97956 3 0.72838 0 0.978911 0.39477 0 0.943150 0.96772 8 0.76439 0 0.760549 0.74606 5 0.75329 0 256 0.42155 3 0.94304 8 0.97105 3 0.90638 7 0.98105 6 0.73573 0 0.984876 0.44597 0 0.952091 0.98418 5 0.77129 8 0.759630 0.82235 4 0.72383 8 512 0.42250 1 0.95206 2 0.98527 0 0.96068 6 0.97907 3 0.76625 8 0.988171 0.50164 0 0.962880 0.98930 7 0.77188 3 0.746005 0.78872 6 0.78970 9 1024 0.42378 9 0.90940 0 0.97617 5 0.95949 0 0.98287 1 0.88110 4 0.990863 0.51403 7 0.966186 0.98677 5 0.76977 9 0.749507 0.80411 8 0.77811 6 Dimensi o n Foetal ECG Glass F ur n a c e Great L a k e s Koski ECGLeleccumMemoryNetworkOceanOcean ShearPacket PGT50 A lp h a PGT50 C D C1 5 Power D a t a Power P l a n t 16 0.58236 6 0.66729 2 0.76361 7 0.78821 1 0.73878 8 0.76707 7 0.361056 0.83872 9 0.45440 9 0.25294 2 0.201408 0.67343 1 0.76344 1 32 0.75446 0 0.80528 6 0.81717 6 0.86844 2 0.80352 7 0.87391 3 0.441143 0.90558 9 0.48506 1 0.32784 6 0.309233 0.73799 7 0.80138 7 64 0.82509 4 0.82723 3 0.85598 5 0.91499 9 0.87770 5 0.94623 7 0.478237 0.94732 1 0.53341 1 0.42988 5 0.378305 0.82087 9 0.87809 5 128 0.87791 2 0.87016 7 0.90765 0 0.94480 5 0.92454 9 0.97281 2 0.546761 0.97041 9 0.56830 9 0.51950 9 0.459351 0.86443 2 0.92977 2 256 0.88818 5 0.87107 9 0.91629 2 0.97294 0 0.94699 5 0.98323 9 0.587142 0.98367 9 0.60610 8 0.61730 5 0.561448 0.91121 2 0.94161 3 512 0.88858 8 0.85820 4 0.92472 7 0.98176 2 0.96439 4 0.99044 7 0.612243 0.98577 8 0.65792 0 0.68272 4 0.594279 0.93749 7 0.95399 4 1024 0.93721 2 0.85040 4 0.98248 8 0.98874 6 0.98021 3 0.99388 6 0.613137 0.98651 4 0.69202 2 0.72140 4 0.678522 0.95269 2 0.96960 4 Dimensi o n Random W al k Robot ArmShuttleSoil TempSpeech Spot E x r a t e s Standard & Poo r Steamgen Synthetic Cont rol TideTongueWindingWoolAverage 16 0.69306 0 0.63865 1 0.84883 7 0.46639 3 0.80283 5 0.66735 1 0.708446 0.64021 5 0.197165 0.68480 9 0.70238 5 0.745298 0.78269 8 0.65353 5 32 0.76272 6 0.67236 0 0.88758 2 0.58346 8 0.81397 2 0.80200 3 0.815125 0.75307 3 0.242413 0.77534 0 0.82186 6 0.833679 0.89775 3 0.72802 9 64 0.85223 4 0.70346 3 0.92737 0 0.60362 9 0.86502 1 0.84558 7 0.898330 0.83477 4 0.357602 0.86674 5 0.86055 7 0.881309 0.93422 7 0.77830 2 128 0.90929 1 0.65738 2 0.96970 1 0.63595 9 0.90771 1 0.91762 2 0.922461 0.93956 8 0.525450 0.90278 1 0.87671 3 0.905980 0.96448 3 0.81901 4 256 0.95135 6 0.65978 0 0.97537 0 0.66348 9 0.91422 1 0.95704 6 0.950578 0.95279 1 0.647056 0.91857 3 0.88326 6 0.912312 0.98000 2 0.84243 5 512 0.97178 8 0.63630 1 0.98953 9 0.66724 9 0.91758 5 0.96517 9 0.968773 0.96321 2 0.673337 0.93247 4 0.91030 5 0.929074 0.99009 1 0.85701 0 1024 0.98134 9 0.67786 5 0.99620 4 0.65422 5 0.91614 6 0.96820 8 0.978736 0.96148 5 0.714639 0.94314 3 0.90970 5 0.957939 0.99606 1 0.87027 5

10. Varying Scaling Factor The following slide shows the effect of varying the range of allowed scaling factors on pruning power. –Note the x-axis indicates the upper bound range of allowed scaling factor. The lower bound range of allowed scaling factor is the reciprocal of the upper bound. For instance, the label 2.0 indicates that the range of allowed scaling factor is between 1 / 2.0 = 0.5 and 2.0. In particular, the label 1.0 indicates that the time warping distance was calculated without scaling. It also implied that the size of the range was not increasing linearly. –However, the important observation is that for all sizes of ranges, a pruning power of over 90% was achieved in nearly three- fourths of the datasets. –For almost all datasets, the pruning powers never dropped below 60%.

12. Varying Scaling Factor We note that vigorously fluctuating datasets are far less common than smooth datasets. –The following slide illustrates this claim by showing the pruning power averaged over all the datasets, as the range of allowed scaling factor changes. For most ranges of scaling factors, the pruning powers achieved are above 90%.

14. Pruning Power vs. Scaling Factor SFEEGERP Data Reality Check ATTAS Ballbea m Buoy Sensor BurstBurstinChaoticCSTRDarwinEarthquakeEeg Evaporato r 1.00.9951550.9885300.9995000.991652 0.99948 3 0.9710440.999601 0.98045 1 0.999443 0.99961 9 0.9985170.9955550.9975950.984193 1.10.5697880.9475830.9981500.982738 0.99539 0 0.9455880.998441 0.69090 6 0.992562 0.99820 2 0.8711370.8331630.8921530.896910 1.20.5106370.9366550.9964210.969948 0.99369 8 0.9178380.997110 0.62807 5 0.984201 0.99546 1 0.8412030.8210580.8807530.852248 1.30.4874030.9257150.9895860.970420 0.99089 3 0.9251520.996125 0.59287 7 0.981369 0.99229 2 0.8091790.7707090.8089650.860910 1.40.3745070.9116370.9792980.959336 0.98253 1 0.8814100.993280 0.50365 9 0.967504 0.98998 2 0.7479590.7093240.7711870.817780 1.50.4237890.9094000.9761750.959490 0.98287 1 0.8811040.990863 0.51403 7 0.966186 0.98677 5 0.7697790.7495070.8041180.778116 1.60.5999080.9549940.9905570.963557 0.99173 2 0.9479080.997250 0.70678 3 0.988641 0.99376 6 0.8684810.8132280.8377580.948679 1.70.5999080.9550040.9905640.963557 0.99187 9 0.9479280.997251 0.70678 3 0.988643 0.99390 9 0.8684810.8132280.8377610.948679 1.80.5950270.9546200.9876760.949513 0.99026 3 0.9436700.996806 0.73864 4 0.988885 0.99346 9 0.8937900.8251440.8341760.950930 1.90.5950270.9546200.9876810.949513 0.99026 3 0.9436710.996809 0.73864 4 0.988886 0.99350 4 0.8937900.8251460.8341760.950930 2.00.5719190.9469460.9861030.948730 0.98804 3 0.9332960.995620 0.71495 4 0.986619 0.98954 1 0.8824680.8057410.8165870.939334 SFFoetal ECG Glass Furnace Great Lakes Koski ECG Leleccu m MemoryNetworkOceanOcean ShearPacket PGT50 Alpha PGT50 CDC15 Power Data Power Plant 1.00.9976240.9949210.9992670.999239 0.99912 2 0.9993520.989543 0.99587 4 0.94993 6 0.9537110.9333910.9974460.999292 1.10.9763550.9386640.9910920.997372 0.99469 0 0.9984410.788691 0.99521 8 0.81196 6 0.8667750.8171160.9799900.991370 1.20.9735170.8958590.9906630.994860 0.99020 9 0.9973470.715996 0.99308 1 0.77029 6 0.7902300.7054620.9733940.989214 1.30.9639910.9128470.9916360.994961 0.99005 6 0.9965920.661433 0.99276 8 0.75494 5 0.8178640.7318990.9634060.982271 1.40.9445340.8710090.9706730.991893 0.98116 9 0.9950020.593253 0.98753 2 0.75174 5 0.7286780.6509730.9563450.970720 1.50.9372120.8504040.9824880.988746 0.98021 3 0.9938860.613137 0.98651 4 0.69202 2 0.7214040.6785220.9526920.969604 1.60.9661990.9899530.9919070.995198 0.98705 1 0.9972800.761301 0.98783 9 0.83141 7 0.8748200.8731530.9754400.985869 1.70.9661990.9902560.9919070.995199 0.98705 4 0.9972810.761301 0.98783 9 0.83141 7 0.8748200.8731530.9755410.986111 1.80.9669430.9894890.9740320.995348 0.98706 5 0.9965160.738819 0.98480 2 0.86325 4 0.8982950.8898210.9584610.982541 1.90.9669430.9894890.9740320.995349 0.98706 6 0.9965260.738820 0.98503 7 0.86325 4 0.8982950.8898210.9586270.982550 2.00.9601830.9879190.9709520.992076 0.98455 2 0.9951830.712417 0.98322 7 0.84526 5 0.8833580.8802980.9517420.968619 SF Random Walk Robot ArmShuttle Soil Temp Speech Spot Exrates Standard & Poor Steamg en Synthetic Control TideTongueWindingWoolAverage 1.00.9993160.9918230.9994720.997178 0.99952 5 0.9967800.998003 0.99846 3 0.977751 0.99884 2 0.9994150.9983950.9991560.991684 1.10.9953310.8535490.9988950.786691 0.97024 3 0.9876560.992618 0.98899 9 0.785640 0.99378 5 0.9859250.9874070.9986100.928805 1.20.9914900.7752360.9983850.692403 0.96129 2 0.9801510.986087 0.98278 9 0.752406 0.98451 7 0.9573320.9824300.9976210.905870 1.30.9898520.7388840.9980830.689417 0.96120 6 0.9771200.985869 0.98474 8 0.711955 0.98329 0 0.9607340.9860880.9972060.897890 1.40.9803210.6832940.9971180.630318 0.91896 3 0.9679300.977566 0.97348 1 0.677345 0.96193 4 0.9147820.9623380.9972770.868613 1.50.9813490.6778650.9962040.654225 0.91614 6 0.9682080.978736 0.96148 5 0.714639 0.94314 3 0.9097050.9579390.9960610.870275 1.60.9863260.8677260.9982390.676847 0.96922 2 0.9816360.988090 0.98078 3 0.819298 0.98512 9 0.9854960.9862750.9977540.927593 1.70.9863280.8677260.9982420.676847 0.96922 3 0.9816440.988091 0.98082 5 0.819298 0.98513 7 0.9854960.9862780.9977580.927619 1.80.9814630.8704260.9974390.687779 0.96611 3 0.9798300.987544 0.97262 2 0.859444 0.97902 7 0.9812150.9810850.9972190.929122 1.90.9814650.8704260.9974400.687779 0.96611 3 0.9798300.987545 0.97262 5 0.859444 0.97902 7 0.9812150.9810850.9973390.929142 2.00.9771630.8499370.9969170.667356 0.95630 3 0.9763850.982182 0.96578 4 0.841658 0.97180 8 0.9762450.9756630.9968760.920468

15. Relative Page Access The following slide shows the relative page access of the datasets versus the length of data. –The relative page access varied significantly from approximately less than 0.2 and up to less than 1.8. –For short data of length 16, the relative page access is almost always larger than 1, suggesting that it is generally not a wise idea to use index for short data. However, as the length of data increases, the relative page access decreases in general, as evident in the slide following next, which shows that the relative page access decreases as the length of data increases. –This is an important result for the proposed index, signaling that the index is likely to perform progressively better as the length of data increases.

17. Average Relative Page Access The following slide also shows that the relative page access approaches 1 when the length of data is between 64 and 128. –This suggests that the use of index should be considered if the length of data is greater than 64. –The fact that about half of the datasets achieved a relative page access below 1 at length 64 and that over 70% of the datasets achieved less-than-one relative page access at length 1024 backed up the above claim.

19. Relative Page Access vs. Length of Data Dimensi o n EEGERP Data Reality C h e c k ATTASBallbeam Buoy S e n s o r BurstBurstinChaoticCSTRDarwinEarthquakeEegEvaporator 16 1.51822 8 1.33865 7 1.24167 5 1.76020 2 1.19783 9 1.34961 5 1.256644 1.45887 5 1.189462 1.30355 5 1.54206 8 1.452640 1.48027 0 1.19976 2 32 1.52894 0 1.33418 0 1.18690 8 1.48384 1 1.17487 8 1.24233 4 1.083856 1.38304 1 1.267108 1.26235 7 1.43533 6 1.318991 1.33140 6 1.12041 3 64 1.56660 9 1.10858 9 0.83041 5 0.90657 8 0.85234 7 1.21532 1 0.763797 1.38175 0 1.155789 0.98324 9 1.21313 8 1.407445 1.18663 3 1.05183 1 128 1.61715 3 0.87560 3 0.48791 7 0.60225 0 0.93610 4 0.98526 1 0.526650 1.18565 8 1.364667 0.59001 4 1.22954 9 1.549319 1.12621 5 0.76140 7 256 1.55351 5 0.95636 7 0.37072 7 0.47338 7 0.69075 9 0.93708 6 0.428182 1.43543 9 1.114336 0.43353 1 1.29117 5 1.428954 0.93681 8 0.78073 7 512 1.46659 9 0.70830 4 0.19108 8 0.25765 5 0.66703 3 0.89446 6 0.395736 1.21857 4 0.868454 0.30995 5 1.23484 8 1.292325 0.84921 1 0.72585 6 1024 1.41603 4 0.93373 3 0.31911 8 0.26880 5 0.57877 5 0.75347 8 0.431923 1.57335 9 0.829318 0.41391 6 0.98994 0 1.347502 1.09414 5 0.90584 8 Dimensi o n Foetal ECG Glass F ur n a c e Great L a k e s Koski ECGLeleccumMemoryNetworkOceanOcean ShearPacket PGT50 A lp h a PGT50 C D C1 5 Power D a t a Power P l a n t 16 1.34068 9 1.28743 3 1.43558 3 1.23788 5 1.22426 1 1.21566 5 1.383371 1.15370 2 0.99275 2 1.48101 2 1.634052 1.32720 3 1.13692 0 32 1.24725 0 1.31182 3 1.20759 0 1.12516 5 1.13867 8 1.09904 5 1.163370 1.02447 5 1.13271 8 1.26021 4 1.316821 1.15076 3 1.23305 1 64 1.03830 4 0.89653 9 0.98863 2 0.84030 3 1.02212 7 0.71525 1 1.195175 0.63168 6 0.98616 0 1.22137 8 1.450676 0.85811 5 0.93092 7 128 1.04863 8 0.58183 6 0.95528 9 0.61901 8 0.71747 6 0.47932 0 1.044484 0.44064 1 1.05840 2 1.25011 4 1.376594 0.79149 6 0.73023 4 256 0.96075 1 0.82988 0 0.83100 2 0.40363 8 0.32735 1 0.30891 8 1.206453 0.28395 2 0.96026 7 1.47184 2 1.263139 0.61871 2 0.50073 6 512 0.93054 4 0.66788 8 0.63037 0 0.44067 5 0.20297 9 0.15741 7 1.127305 0.20598 2 1.01980 1 1.37478 3 1.382482 0.47111 8 0.46236 1 1024 0.97912 2 0.65537 6 0.23196 7 0.66591 6 0.15457 8 0.18097 8 1.139579 0.16943 8 1.25458 7 1.40560 0 1.425001 0.46577 2 0.44483 7 Dimensi o n Random W al k Robot ArmShuttleSoil TempSpeech Spot E x r a t e s Standard & Poo r Steamgen Synthetic Cont rol TideTongueWindingWoolAverage 16 1.37664 2 1.68291 3 1.23489 1 1.57379 2 1.39866 9 1.29333 8 1.345880 1.36076 7 1.365710 1.39070 4 1.30882 9 1.201163 1.12837 8 1.34037 6 32 1.34945 1 1.53253 8 1.10473 0 1.49370 1 1.49631 0 1.27106 3 1.329950 1.15971 7 1.355713 1.38589 8 1.37040 7 1.350632 1.14021 6 1.26657 0 64 1.00504 3 1.58803 3 0.74466 7 1.21275 3 1.23051 0 0.85325 3 0.998204 0.96104 3 1.321875 1.13937 8 1.31153 4 1.131427 0.93856 2 1.06016 4 128 0.78838 6 1.46033 7 0.49691 2 1.06751 4 1.34761 7 0.63316 3 0.744228 0.51966 7 1.186771 1.05625 3 1.35330 3 1.047924 0.80032 7 0.92376 5 256 0.42156 6 1.29998 8 0.19513 4 1.06339 8 1.25982 3 0.36124 1 0.352205 0.40780 6 0.864809 0.82194 2 1.40271 0 1.043574 0.49883 8 0.80669 9 512 0.45772 5 1.58526 4 0.21427 3 1.16753 9 1.26739 3 0.30669 5 0.358125 0.54005 4 0.927356 0.74864 6 1.22356 5 0.923770 0.30635 2 0.74113 5 1024 0.25804 4 1.52800 5 0.08830 1 1.09723 5 1.16828 6 0.22810 4 0.169561 0.33132 0 0.568785 0.76921 5 1.22067 0 0.881667 0.13293 5 0.72293 2 Relative Page Access – Raw Numbers

20. Query Time The following slide shows the actual running time of the range queries as calculated by the difference between two calls to gettimeofday before and after the queries. –The time is averaged over all 50 queries performed for each length of data of each dataset. –It shows that the query generally runs very fast. All queries completed within a fraction of a second for all datasets of length 16 and 32, and all queries completed in the magnitude of minutes. –Note the logarithmic scale in both axes.

22. Average Query Time The following slide shows the running time averaged over all datasets. –It suggested that most queries actually completed well within a minute, even for the larger length of data. And even for the largest length of data, queries completed in 560 seconds on average. –Recall from previous slides that linear scan perform better for datasets of shorter lengths; however, as the following slide shows, the query time for those datasets is actually not significant anyway. –Moreover, the proposed index can significantly reduce the query time for datasets of longer lengths. Combining both advantages, our proposed index is capable as an all-round solution suitable for datasets of all lengths.

24. Query Time vs. Length of Data Dimensi o n EEGERP Data Reality C h e c k ATTASBallbeam Buoy S e n s o r BurstBurstinChaoticCSTRDarwinEarthquakeEeg Evaporato r 160.0089060.0089910.0078520.0278200.0085240.0076870.0102440.0085370.0079470.0110910.0103370.0085090.0081300.006077 320.0346020.0275480.0509980.1402720.0208440.0258290.0254150.0239060.0233030.0227330.0219010.0428870.0329260.017459 642.8576231.2052250.7215541.6938150.5644201.7748760.7834521.3840061.0493000.7310080.9288252.3454601.8257440.926490 12818.4977086.2005651.9998905.3161193.6021365.5364491.8991899.18951110.8178022.1570835.65065816.7379718.2003565.450292 25672.65350320.0722626.09093132.63387014.10298146.9327999.18999045.07391330.0449767.94138826.60310475.27561642.68375827.296181 512317.31845262.41701817.06726340.47781852.518124 154.42589 2 26.894878 204.29439 6 78.94824321.008843117.051126552.629948 106.35417 2 138.26356 1 1024 1245.41442 2 604.96168491.006146 106.82902 5 239.95185 7 509.28941 4 161.670165 1251.9276 3 1 279.658997 130.97699 2 448.4496662228.000279 1231.3625 3 9 825.87321 8 Dimensi o n Foetal ECG Glass F ur n a c e Great L a k e s Koski ECGLeleccumMemoryNetworkOceanOcean ShearPacket PGT50 A lp h a PGT50 C D C1 5 Power Data Power P l a n t 160.0122310.0079720.0075450.0092040.0078990.0135360.0082370.0094360.0100560.0054250.0083880.0114040.0083280.007762 320.0220820.0188740.0207690.0297710.0599500.0443970.0288540.0480080.0404050.0327460.0287600.0208380.0275460.026521 641.6666700.8092960.9791230.6203421.8714190.6930612.6804750.5000410.6502381.5536501.4842771.8194460.9868590.821533 12810.4002143.2014898.5538282.8019508.9661951.74924912.9503481.7813161.7896178.6195179.96643011.9327524.1261294.021875 25623.34804640.49028632.6900457.02824010.1470135.16369684.2706226.2753555.07575727.07833692.25856854.40475414.70836510.431865 51298.134765124.507538162.57611843.74687616.63579615.147429254.14985614.82258014.385844 113.62204 7 287.367275304.58696339.68193932.909189 1024352.864132467.268620149.532524 234.98443 9 50.04271349.6467971211.57194464.28668760.260655 917.93398 5 2013.34924 8 1313.071738 207.55748 4 148.65534 9 Dimensi o n Random W al k Robot ArmShuttleSoil TempSpeech Spot E x r a t e s Standard & Poo r Steamgen Synthetic Cont rol TideTongueWindingWoolAverage 160.0079860.0096210.0139350.0096820.0078060.0102970.0080180.0092800.0096460.0099570.0076090.0097570.0077340.009498 320.0256150.0454870.0227710.0343440.0252840.0215460.0363220.0264980.0257370.0280080.0281740.0257200.0209870.032357 641.5478342.5950870.6430381.6729571.4983410.8473080.8293221.7472612.4032640.9424310.9396330.8556320.9999611.303666 1283.47950117.8172251.85367411.7107627.5192653.1579553.3030262.63213613.3628634.6675847.5568164.6312063.4445056.762272 2568.90137568.1452563.09377659.23178637.1312636.5175466.34344317.15533139.37082921.75384348.24318025.4538888.87039529.760442 51236.814731375.71248914.908316 325.89924 0 148.18598 8 32.36652725.86497156.191334308.37005393.215671145.99387590.09658423.387428 124.12076 0 102485.938851 2033.80506 1 24.773419 1516.4049 5 7 662.35430 3 101.09487 8 64.856889 149.81725 6 408.384121 312.71638 0 704.605891287.20759338.630299 560.65825 0 Query Time – Raw Numbers