Microsoft Word SoccerPaper5.docx

Transcript

1 Prediction in Soccer using Shot “Quality vs Quantity”: Improved Strategic Features from Spatiotemporal Data and Iain Monfort, Peter Carr Patrick Lucey, Alina Bialkowski, Mathew Matthews Disney Research Pittsburgh, PA, USA, 15232 [email protected] Email: Abstract In this paper, we present a method which accurately estimate s the likelihood of chances in soccer season of using strategic features from an entire pla yer and ball tracking data taken from a professional league . From the data, we analyzed the spa tiotemporal second window for before a shot of play ,000 shots. From nearly 10 patterns of the t en - game phase importan kick, - t (i.e., corner, free our analysis, we found that not only is the strategic features such as defender proximity, play, counter attack etc.), the - open unding players, speed of play, coupled with the shot location interaction of surro play a goal. Using our an impact on determining the likelihood of a team scoring spatiotemporal strategic features, we can accurately measure the likelihood of each ify the efficiency of each team and their strategy. shot. We use this analysis to quant 1 Introduction In the 2014 FIFA World Cup in Brazil, arguably the most memorable match was when Germany hen analyzing ame the shooting statistics for this g - However, w blitzed Brazil in the semi - final 7 1. Brazil actually had more shots and shots on target (18 vs 14 and 13 vs 12 respectively) which flect the sheer dominance that Germany had does not re . In soccer, it is well known that not [1] all shots are created equally, but in this paper we ask the question “how can we quantify the value of a data?” player tracking shot directly from t to consider is the proximity of the An obvious starting poin – shot location to the goal the closer the shot to the goal the more likely it will result in a goal However, additional contextual features such as “space” (i.e., the distance from . (see Figure 1) the defender), and The number of defenders between the shot and goal play an important role. p osition of other attackers their motion paths also give important cues on the quality of shot (as well as uncover how teams get open shots). - fine grained These features can only be derived from g data. player trackin . (Left) Shots the probability distribution of all shot locations, (Right) Shows the probability distribution of 1 Figure shot locations then resulted in a goal 2015 Research Paper Competition Presented by:

2 Counter-Attack Corner Open-Play Free-Kick Penalty Set-Piece Figure (red team has possession and is 2 . Example plays which represent the different match contexts that shots occur attacking left to right). - context also plays an important M . For this atch factor in determining the likelihood of a goal the shots into six different match - context: i) open - play (possession in the work, we partitioned - - end to the other), iii) corners, attack (players break quickly from one forward third), ii) counter - kick (shot on goal from a free kick), and vi) set - pieces iv) penalties, v) free ( a cross that comes - into the box from a free - kick) – visual examples are shown in Figure 2. I n Table 1 it can be seen that a team is more likely to score on a counter attack compared to a normal possession (and of - course, a penalty) . Additionally, a team is more likely to score from a normal possession than a free - kick . In terms of corners, the shot/goal ratio of around 9% appears to represent a reasonable chance, but considering that only a small portion o f corners result in a shot , corners tend to be r ather inefficient which backs up previous work [ 2 ]. Game Context: Open - Play Counter Attack Corners Penalties Free - Kick Set - Pieces 39 ) 39 ( ) Number (Goal) 6467 ( 534 ) 1116 ( 166 ) 1115 ( 100 88 94 (67) 5 3 (2 6 ) % 14.87 % 8.97 % 8.26 71.3% 4.82 % 10.05 % Average Shot/Goal Table 1 . Shows the number of shots and goals for the various shot contexts. In this paper, we present a method which can accurately estimate the likelihood of chances in using strategic features from a seasons worth o f player and ball tracking data from a soccer professional league fro m Prozone [ 3 ]. The league we analyzed had 20 teams and played each team home and away. Due to sensitivities of the league, we anonymized the identity of the league From the data, we analyzed the spatiotemporal and teams - as such we labeled them A - T. second window before a shot of . In this data, the spatial location patterns of the t en - 9732 shots given at 10 frames per second, and the spatial location and time - stamp of ball of players are 53 27 games ( events are given. In the season we analyzed, we used 3 games were omitted). Due ayer to the constant changing of player role, our recent work of aligning mu lti - agent pl - grained team - ] enabled us to craft strategic features which capture fine trajectories [4 7 dynamics, which we then fed to a Conditional Random Field (CRF) [8] to spatiotemporal As soccer is estimate the likelihood of a team scoring from a given chance . ultimately decided by sho ts and goals, our approach analyzes teams as a function of both the quantity and quality of 11 - ], however, our approach uses ve been applied to basketball [9 chances . Similar approaches ha tendencies instead of individual attributes. strategic features which incorporates team 2015 Research Paper Competition Presented by:

3 2 Quantifying Goal Likelihood % of all their shots. A approximately 9.6 On the season we analyze d , on average a team will score naïve method would be to assign this estimate for all shots which woul d lead to a large average Table error of 0.1745. However, knowing the match - context in which the shot was taken (see better estimate we can form a , ) 1 which reduces the average prediction error down to 0.1662. here are many other features that should be incorporated Clearly, this is not satisfactory either as t As can be seen in Figure 1, the shot location also to give a better estimate of goal likelihood. plays an important role in estimating if a goal is going to occur so if we further condition the For anyone who has watched like lihood of the spatial location we can reduce this further. fine - soccer, however, these are still very coarse measures and are devoid of grain context. For - /her chances are per, his example, in a counter - attack, if a player is one kee on - one with the goal - increased. Or given space on top of the 18 yard box, a player is more likely to score given space grained player - from the defenders. The only means of getting such features is from using fine m this information. tracking data and crafting features fro Using this data, not only can we obtain c elements such as the motion of important spatial and action data, we can also include strategi surrounding players and the structure of the defending team. In the following subsections, we be how we captured these semantic and strategic elements. descri Defender 2.1 Proximity proximity effects the decision that will be made as well as the Having a defender in close execution. As the major goal of a defender is to protect the goal, the orientatio n of their position relative to their goal needs to be captured. The way we determined defender proximity is by first he shot and the goal (see Figure 3 checking if any defending players were in the area between t ). If they were, we calculated the Euclidean distance between the shot location and defender. If a defender was not within this area, we gave this distance a negative value. In open - play, when a (i.e., not within the shaded space) the defender was not goal - side when a shot occurred % , compared to 11.59% oal increased to 23.18% of where (p<0.00001) 7.49 likelihood of a g shots were “open” . Similarly for counter - attacks, the likelihood of a goal increased to 18.44% (p<0.01) where 40.32% of shots were “open” . Additionally, for - compared to 12.46 % counter attacks getting a n open shot occurred more often which makes sense as more space is created in a counter attack. Of course, this also dep ends on the distance from goal. We also used the rmine if a defender was within this number of defenders in this area as a feature as well. To dete area, we polygon calculation. - in - defined the vertices of the triangle and then used standard point To capture how much pressure the shooter is under, we devise a defender proximity descriptor which first counts . 3 Figure how many defenders are in the space between the goal and the shooter and then we get the distance between the shooter and the defenders (red ar e the attacking players and blue are the defenders). 2015 Research Paper Competition Presented by:

4 2.2 Defensive Formation/Structure The shape and defensive structure plays an important part in the likelihood of a goal. Quantifying team structure is difficult however, but our recent work in this area has allowed us to craft features to measure such behavior [4 - 7] . Vital to this estimation is to determine the “role” of each player within the formation. This is done by finding the permutation of the raw which minimizes the distance to the base template. Once this cost matrix is location points cal c to make the assignment of role to each player. ated, we use the Hungarian algorithm [12] ul Once this is determined, w he defensive the following features: i) the distance between t ed e craft line, ii) the distance between the back - line and the midfield line, iii) the number of defensive role - , swaps . and iv) the number of attackers in - front of the defensive center 2.3 Attacking Features In th is subsection, we describe some of the attacking ata. features we extracted from the d Important factors which we wanted to extract where, was it a long pass , cross, dribbling and which lead to the shot on goal Additionally, . taking on players, or pressing (causing turnovers) layer who gave the incoming pass/cross also plays an important role as it the space of the p suggests the quality could be potentially higher. The pace of the players moving and how the attacking team moves relative to the opposition was also captured in our feature set. Strategic Features for each Game Expected Goal Value using 2.4 Context - Given the game - context and the various spatio temporal features, we can estimate the likelihood of each shot using logistic regression. We call this estimate Expected Goal Value (EGV), which is similar to other approaches used in basketball [9 11]. Approaches such as these have - also been used in soccer, but have not included player tracking data – As the just ball - event data. game context clusters and - to distinct game context is important, we first partition the examples in - fitting, we used - context. To avoid over - r for each of these 6 game n individual regresso learn a regularization and we divided the examples into a train/tes t set. Using the features we show in Table 2 how the average err or of our prediction lowers. Context + Factor Average Shot - Context + Location + Location + + Context Context Defending + Attacking Likelihood Location Defending 0.1439 0.1554 0.1545 Average 0.1745 0.1662 Error 2 Table . Showing the residual when we use different methods to estimate the likelihood of a shot resulting in a goal. 2015 Research Paper Competition Presented by:

5 Team and Game nalysis 3 A 3.1 Team Efficiency Ratings (Season wide analysis) ood of a goal, allows us to do deeper analysis Having the ability to better estimate the likelih First of all, we can evaluate the which may help unlock characteristics or traits of teams. efficiency of a team’s performance in terms of offense and defense and compare them to the rest his as a starting point, we can drill down further to check how efficient each of the teams. Using t team is in terms of different match context. Let us first analyze the attacking performance across the season (due to some missing matches, the overall statistics here may not m atch the complete performance is shown in Table 3 season). The . GOALS ID Average SHOTS ID SHOTS GOALS EGV Average EGV Error Error 514 58 51.91 7.89 371 34 35.64 4.79 A A B 434 46 39.47 5.85 B 620 62 59.31 8.47 C 68 63.62 9.57 594 443 35 38.42 5.12 C D 562 46 50.4 6.95 D 415 37 41.68 5.87 440 42 42.57 6.21 E 604 58 60.01 9.09 E F 65 65.85 9.28 F 407 38 37.45 5.01 694 G G 59 62.85 9.65 593 353 26 31.19 4.12 H 514 71 57.21 10.35 H 451 38 35.16 4.46 474 41 40.16 5.33 I 458 59 47.13 7.96 I J 416 39 41.45 5.78 J 533 56 51.63 7.76 6.53 47.61 48 547 K K 447 26 35.95 3.93 L 33 34.3 4.66 364 614 62 59.02 8.73 L M 464 42 44.12 6.01 M 389 50 44.87 8.02 N 338 29 34.72 4.54 N 447 35 40.2 4.76 416 39 38.93 5.38 O O 592 50 61.33 8.16 P 467 45 44.95 6.86 P 523 49 53.13 7.29 6.13 36.3 41 Q 611 57 50.13 6.98 Q 344 458 43 46.39 6.4 0 R R 529 46 52.38 7.08 S 479 40 43.34 5.86 S 576 45 45.41 5.76 64 T 458 44 38.44 5.28 T 521 7.68 48.85 offensive Table All Shots: (Left) The . shooting statistics , and (Right) defensive statistics for every team in the league. 3 Columns 5 and 6 give the expected goal value and the average error per prediction. The rows highlighted in bold highlight teams where their goals is signi ficantly different than their EGV. In terms of offense (left), taking into account the error of the estimate, most teams scored within their expected range with the exception of two teams. Team H were very efficient, scoring 71 goals (but with an expected goal value of approximately 57±10 goals). Even with the maximum - error, a difference of 4 goals is quite significant. On the other Team K hand, only scored 26 goals, which was differe nt from their EGV of 36±4. As both teams finished at either ends of the table, the quality of strikers may suggest the difference between actual goals scored and their th conceding EGV. In terms of defense, similar patterns emerge with Team I eir 59 goals when EGV was around 47±8. Team T also game up more goals then expected, with a EGV of 49±8, when they actually gave up 64 goals. Poor goal - keeping, excellent strikes by the opposition or a combination of the two could have caused t his. Team O on the other h and, were expected to give up 61±8 goals, but only conceded 50 goals which maybe due to the inverse of the previous example. Performance may vary for various match contexts too. In Table 4, we show the EGV for the play. When we focus on this particular match - various teams based on shots just from o pen - Teams H and K context, still have a big difference between actual goals and EGV, but three is could be due to the fact scored more goals then expected (th Team Q other teams do as well. Teams N and E some incredible goals that season). score they had a player who underachieved 2015 Research Paper Competition Presented by:

6 Team N though, which may suggest the lack of quality for those teams. In terms of defense, only play with their EGV being higher at 25.5±2.8. conceded 17 goals in open also conced ed Team P - conceded much more than expect for shots in open Teams I and T much less than expected. - play. SHOTS EGV Average ID GOALS EGV Average GOALS SHOTS ID Error Error 223 31.01 4.66 A 35 16 17.18 2.08 358 A 281 26 23.38 3.43 B 427 37 34.05 4.8 0 B 37 36.16 5.35 C 309 22 26.19 3.33 C 401 390 31 33.25 4.66 D 258 18 22.08 2.91 D 297 26.19 3.57 E 420 33 33.28 4.56 E 20 F 468 38 37.85 5.12 F 267 18 21.38 2.66 G 30 32.87 4.7 0 210 16 16.21 2.05 G 371 332 39 27.52 4.27 H 297 23 20.44 2.51 H I 26 24.42 3.19 I 294 38 26.21 4.53 326 J J 20 23.31 3.04 276 368 34 30.07 4.33 K 290 14 19.49 1.87 352 30 25.99 3.46 K 225 17 18.67 2.57 L 420 40 35.6 0 5.12 L M 28 27.34 3.67 M 245 25 23.51 3.86 328 14 18.29 2.28 N 315 17 25.5 0 2.78 N 214 249 19 18.77 2.42 375 26 33.1 0 4.43 O O P 333 35 30 4.63 P 353 23 30.52 3.91 Q 411 37 29.94 4.14 Q 230 25 21.73 3.55 R 287 19 21.63 2.6 0 R 347 25 29.77 3.97 S 26 26.72 3.6 0 334 S 420 33 30.45 3.94 T 296 23 21.78 2.73 T 337 35 25.32 3.71 4 Open Play: (Left) The offensive shooting statistics , and (Right) defensive statistics Table for every team in the league. . The rows highlighted in bold highlight Columns 5 and 6 give the expected goal value and the average error per prediction. teams where their goals is significantly different than their EGV. SHOTS GOALS EGV Average SHOTS Average ID ID GOALS EGV Error Error A 63 10 9.53 2.13 A 56 7 8.07 1.57 9.92 1.88 8.1 8 47 B B 57 10 2.27 C 8 8.96 1.79 C 51 7 7.3 1.64 56 D D 5 4.83 0.97 34 53 10 8.53 1.98 E 47 8 9.21 2.19 E 70 11 11.27 2.69 F 88 15 15.31 3.46 F 36 5 6.14 1.25 G 88 13 13.08 2.73 G 60 5 7.62 1.29 82 16 13.85 3.59 H 59 8 6.95 1.31 H I 59 9 7.18 1.48 I 57 9 9.87 2.39 38 4 6.24 1.19 J J 58 14 9.25 2.59 K 52 6 7.38 1.54 K 53 6 8.19 1.57 88 L 2.01 11.52 8 L 35 3 4.06 0.67 44 5 5.47 0.84 M 42 11 7.62 2.09 M N 35 5 5.12 1.06 N 38 4 5.06 0.93 59 8 7.04 1.49 O O 61 11 12.5 2.93 P 51 6 8.4 1.82 P 51 8 8.88 1.71 77 4 2.05 7.84 Q 11 38 13.56 2.99 Q R 57 10 9.85 2.11 R 57 9 9.23 1.81 7.29 S 1.08 S 35 4 5.34 1 .00 5 64 T 59 10 8.01 1.7 0 T 77 16 11.11 2.61 shooting statistics , and (Right) defensive statistics offensive for every team in the Table 5 . Counter Attack: (Left) The The rows highlighted in bold league. Columns 5 and 6 give the expected goal value and the average error per prediction. highlight teams where their goals is significantly different than their EGV. - between actual goals scored and conceded for counter In Table 5, we show the difference attacks. There is no enormous gaps between the actual goals and EGV apart from Team J who Team T defense which game up 14 goals, when they should have only gave up 9.2±2.6 and were expected to only give up 11.1±2.6. gave up 16 goals when they 2015 Research Paper Competition Presented by:

7 Individual Game Analysis 3.2 where the statistics do not tell the Circling back to our original example of Brazil vs Germany in this subsection we show that full story of a match to give a , our analysis can also be used y ”. What we mean by that is soccer better indication on whether a team was “dominant ” or “luck and is still rather random by nature due to the fact that goals are sparsely occurring events keeper having a bad day , or all shots for a particular team being outliers can occur such as a goal - successful. We show some examples in Table 6. In the top 3 examples, we show three matches where teams with significantly less shots won (the first two by large margins), but our EGV mation of dominance. In the remaining 6 examples, we show measure gave a better approxi matches where the dominant team did not win despite having the better chances. Over the season these tend to cancel each other out, but in terms of individual data points this can give a cation of how the match was played. better indi Example 2 Example 3 Example1 M S Teams Teams K P Teams I S Shots 17 11 Shots 22 14 Shots 18 12 5 Goals 3 Goals 0 0 Goals 0 1 EGV 1.50 2.89 EGV 1.39 2.02 EGV 0.83 1.54 Example 5 Example 6 Example 4 Teams I O Teams C M Teams O L 19 15 Shots 7 19 Shots Shots 17 9 2 Goals 3 Goals 2 0 Goals 3 0 0.74 1.37 EGV EGV 2.14 0.66 EGV 1.65 1.65 Example7 Example 8 Example 9 Teams F B Teams F R Teams F N 5 5 18 Shots Shots 29 10 28 Shots 0 1 3 Goals Goals 2 Goals 0 0 EGV 0.07 2.25 EGV 2.66 0.75 EGV 2.87 0.53 Table 6 . Examples of matches using the EGV measure to given a better idea of how the match was played and which team dominated. Quantifying Chances: 4 Examples It is one thing to have a reasonable model, but if the predictions do not look “reasonable” then there is a good chance something is going wrong. In this section, we visualize some of our wn in Figure 4. In the top predictions to show that it passes the “eye - test”. Examples are sho - left, a play which has the left winger controlling down the left uncontested and then slotting the yard box results in a chance of 70.59%. In the - ball between the back four to a player in the six ccurred with the ball being crossed to the striker with a second example, a similar break o kick - defender in close proximity which reduced the goal likelihood. In the third example, a free - which was taken and was parried by the goal keeper had a chance of around 50% ending up as a e., for every 2 times you see that occur, one will go in). The fourth example shows a goal (i. box giving a likelihood of 46.10%. However as this - corner which results in a shot in the six yard would normally be taken by a goal t in this instance is low. keeper the likelihood of getting a sho - The remaining examples show low percentage shots often occur when the location is outside the box and a defender is in the way. We didn’t show penalties kicks, as these have little variance in terms of strategic factors. 2015 Research Paper Competition Presented by:

8 70.59% 49.75% 53.14% 18.28% 46.10% 11.12% 6.46% 9.74% 5.47% 4.81% 4.90% 4.44% 4 . Examples showing the goal likelihood from various examples (red team has possession and is attacking left to Figure right). 5 Summary In this paper, we present ed a method which accurately estimate s the likelihood of chances in soccer using strategic features from an entire season of player and ball tracking data taken from a professional league . From the data, we analyzed the spa tiotemporal patterns of the ten - second our analysis, we found that not only window of play before a shot for nearly 10 ,000 shots. From is the game phase important (i.e., corner, free kick, open - play, counter attack etc.), the strategic - features such as defender proximity, interaction of surro unding players, speed of play, coupled with the shot location play an impact on determining the likelihood of a team scoring a goal. Research Paper Competition 2015 Presented by:

9 REFERENCES Final Match, Brazil vs Germany Match Statistics, - [1] 2014 Fifa World Cup Semi . http://www.fifa.com/worldcup/matches/round=255955/match=300186474/statistics.html [2] C. Anderson and D. Sally, “The Numbers Game: Why Everything You Know Ab out Soccer is Wrong”, Penguin Books, 2013. [3] Prozone. http://www.prozonesports.com [4] A. Bialkowski, P. Lucey, P. Carr, Y. Yue and I. Matthews, “Win at Home and Draw Away: hting the Differences in Home and Away Team Automatic Formation Analysis Highlig Behaviors”, in MIT Sloan Sports Analytics Conference (MITSSAC) , 2014. [5] P. Lucey, A. Bialkowski, P. Carr, S. Morgan, Y. Sheikh and I. Matthews, “Representing and Discovering Adversarial Team Behaviors using Pla International Conference of yer Roles”, in the 2013. Computer Vision and Pattern Recognition (CVPR), Scale Analysis - [6] A. Bialkowski, P. Lucey, P. Carr, Y. Yue, S. Sridharan and I. Matthews, “Large International Conference of Data n the of Soccer Matches using Spatiotemporal Tracking Data”, i , 2014. Mining (ICDM) [7] A. Bialkowski, P. Lucey, P. Carr, Y. Yue, S. Sridharan and I. Matthews, “Identifying Team ICDM Style in Soccer using Formations Learnt from Spatiotemporal Tracking Data”, in the , 2014. atial and Spatiotemporal Data Mining (SSTDM) Workshop on Sp [8] J. Lafferty, A. McCallum and F. Pereira, “Conditional Random Fields: Probabilistic Models International Conference on Machine Learning for Segmenting and Labeling Sequence Data”, in (ICML) , 2014. Grained Spatial Y. Yue, P. Lucey, P. Carr, A. Bialkowski and I. Matthews, “Learning Fine - [9] Models for Dynamic Sports Play Prediction”, in the International Conference of Data Mining (ICDM) , 2014. , “Factorized point process intensities: A [10] A. Miller, L. Bornn, R. Adams, and K. Goldsberry , International Conference on Machine Learning (ICML) spatial analysis of professional basketball,” in 2014. [11] D. Cervone, A. D’Amour, L. Bornn and K. Goldsberry, “POINTWISE: Predicting Points and Valuing Decisi ons in Real Time with NBA Optical Tracking Data”, in MIT Sloan Sports Analytics Conference (MITSSAC) , 2014. H. W. Kuhn, “The hungarian method for the assignment problem,” ] [1 Naval Research Logistics 2 97, 1955. – 2, pp. 83 - , vol. 2, no. 1 Quarterly 2015 Research Paper Competition Presented by:

Related documents

Directory

Directory

Intel International Science and Engineering Fair Finalist Directory 2018

More info »
2019 FYE online

2019 FYE online

First-Year & Common Reading 2019 New and Recommended Books Search Press Seven Stories Press Shambhala Publications Smithsonian Books Soho Press Steerforth Press Titan Books Verso Books Watkins Media L...

More info »
D2 Summit OC 19

D2 Summit OC 19

THE D2 SUMMIT - WILD CARD - ORDER OF COMPETITION TIMES FOR FRIDAY MAY 10, 2019 - ESPN WIDE WORLD OF SPORTS ® (Hp Field House) 12 Minute 12 Minute Competition Competition Division Check-In Warm-up Warm...

More info »
Microsoft Word   Rulebook19

Microsoft Word Rulebook19

4HHSG 05 Florida 4-H Area & State Horse Shows North East West South Official Rules 2019 Florida Cooperative Extension Service IFAS, University of Florida, Gainesville Nick Place, Dean and Director for...

More info »
EY NextWave Data Science Challenge Terms and Conditions

EY NextWave Data Science Challenge Terms and Conditions

EY NextWave Data Science Challenge Terms and Conditions 1 April 2019 ORGANISERS OF COMPETITION The EY NextWave Data Science Challenge 2019 (the “Competition”) is conducted by EYGS LLP 6 More London Pl...

More info »
agency search white paper

agency search white paper

ANA/4A’s Guidelines for Agency Search Marketers and their agencies are looking for high-performance relationships based on mutual trust. Steps that can help marketers and agencies reach that productiv...

More info »
625137

625137

2018-19 Nebraska All-Sports Record Book - Nebraska Communications Office -

More info »
art in state buildings handbook

art in state buildings handbook

Art i n State Buildings Program Handbook for State Agencies Florida Department of State, Division of Cultural Affairs on Arts and Culture Florida Council Return to Program Info Handbook • Welcome • Ab...

More info »
ED581422

ED581422

NNSTOY Social Justice Book List AUGUST 2017 Edited by Katherine Bassett, Brett Bigham, and Laurie Calvert Contributors: Revathi Balakrisnan, Katherine Bassett, Brett Bigham, Nate Bowling, Chad Miller,...

More info »
NNSTOY Social Justice Book List

NNSTOY Social Justice Book List

NNSTOY Social Justice Book List AUGUST 2017 Edited by Katherine Bassett, Brett Bigham, and Laurie Calvert Contributors: Revathi Balakrisnan, Katherine Bassett, Brett Bigham, Nate Bowling, Chad Miller,...

More info »
Microsoft Word   2018 PIA Handbook Final.docx

Microsoft Word 2018 PIA Handbook Final.docx

THE OFFICE OF THE ATTORNEY GENERAL OF TEXAS PUBLIC INFORMATION ACT Handbook 2018

More info »
Chapter0Solutions 000

Chapter0Solutions 000

State the domain and range of each relation. Then determine whether each relation is a function. Write yes or no . {(2, 7), (3, 10), (1, 6)} 1. SOLUTION: The domain is the set of x - coordinates. The ...

More info »
private service provider implementation manual

private service provider implementation manual

Private Service Provider IMPLEMENTATION MANUAL

More info »
City Bidding Book Washington State

City Bidding Book Washington State

CITY BIDDING BOOK Washington State Last Updated: July 2018

More info »
Slide 1

Slide 1

The Search Process Overview Office of Diversity, Equity, and Access August 2017 1

More info »
S133 Kansas 4 H Horse Show Rule Book

S133 Kansas 4 H Horse Show Rule Book

Kansas 4-H Horse Show Rule Book Revised Spring 2018 www.kansas4-H.org Kansas State University Agricultural Experiment Station and Cooperative Extension Service

More info »
The End of Man Zylinska

The End of Man Zylinska

The End of Man: A Feminist Counterapocalypse [author ’ s manuscript] Joanna Zylinska 1

More info »
Percent for Art Guidelines

Percent for Art Guidelines

PERCENT FOR ART GUIDELINES City of Portland and Multnomah County Revised October 2006 Throughout history art has been instrumental in creating unique public places that have yielded physical, social a...

More info »