{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Introduction\n\nThe punt play is a unique play in football with several very specialised positions, in this analysis I am focusing on the Gunner and the Vise and their effect on the Returner and the outcome of the punt.\n\nQuick summary of the roles\n- Gunner \n    - Goal: To get down field as quickly as possible to force a fair catch or minimise return yards\n- Vise \n    - Goal: To slow down or prevent the Gunner from reaching or affecting the returner\n\n\nAim of this analysis:\n- Identify who are the best gunners in the league\n- Identify what attributes are common in the best gunners\n- Identify what tactics can be introduced to optimise coaching or technique decisions","metadata":{"execution":{"iopub.status.busy":"2022-01-05T11:37:19.096796Z","iopub.execute_input":"2022-01-05T11:37:19.097201Z","iopub.status.idle":"2022-01-05T11:37:19.107477Z","shell.execute_reply.started":"2022-01-05T11:37:19.097147Z","shell.execute_reply":"2022-01-05T11:37:19.106399Z"}}},{"cell_type":"markdown","source":"# 1: Investigating the battle between Gunners vs Vises\n\nThe first decision which needs to be made in this analysis is to define when and how a gunner has won or lost their battle with the vise. \n\nThere are various ways in which you could define them:\n- Final result of the punt e.g. Returned, Fair Catch\n- Events happened during the punt e.g. tackles\n- When the gunner is closer to the returner than his corresponding vise\n\n\nOut of the options I chose the distance between the gunner and the vise to the returner as this is an easily measurable statistic and can be attributed individually to a player. The final result of a punt cannot be attributed individually as it is a team effort, and events which happen cannot be easily measured as it excludes the events not recorded e.g. pressure on returner.\n\nAt each frame I calculated the distance between the gunner and the returner and the distance between the marking vise (or vises if multiple guarding one gunner) and the returner. At each frame if the gunner was closer than the vise or both vises it was counted as the gunner winning that frame as they have gained a favourable position and in control.\n\nSee example below for what it looks like graphically","metadata":{"execution":{"iopub.status.busy":"2022-01-05T11:37:19.108698Z","iopub.status.idle":"2022-01-05T11:37:19.109203Z","shell.execute_reply.started":"2022-01-05T11:37:19.108992Z","shell.execute_reply":"2022-01-05T11:37:19.10901Z"}}},{"cell_type":"markdown","source":"![](https://i.imgur.com/Fd025yX.png?1)","metadata":{}},{"cell_type":"markdown","source":"# 2: Impact gunners winning has on the punt event\n\nIn the above section we defined at each frame when the gunner was winning and losing. To take this further we aggregated this at a play level by identifying the first frame where the gunner was winning the matchup and compare it to the result of the punt play.\n\nFrom the data below we can see that the longer it takes for the gunners to win their matchup the more likely a play is to be returned. There is a threshold around frame 50 where the likelihood of return happening rapidly increases and the likelihood of a fair catch or downed decreases.\n\nIn real terms this means the gunner needs to win their matchup within 4 seconds (first 10 frames in the tracking data is pre snap) to maximise the probability of a fair catch and minimise the chance of a returned punt.","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/dExI66Z.png)","metadata":{}},{"cell_type":"markdown","source":"# 3: Rankings all gunners & attributes for success\n\nNow that we know how quickly a gunner has to beat their matchup by we can apply that metric to all the starting gunners in the NFL over the past three years. As gunners are sometimes have two vises lined up against them I have separated their success rate against 1 vise and success rate against 2 vises.\nNote due to time constraints this does not factor in tackle % when required or skill at converting touchbacks into downed punts.\n\nDane Cruikshank over the last three years has been the most successful overall gunner with a win rate of 78% vs 1 Vise and 33% vs 2 Vises.\n\nAvonte Maddox has been dominant when facing 1 Vise with a win rate of 88%, while he has had a strong comparative success rate of 22% vs 2 Vises I would recommend that he should be double teamed on punt plays.","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/1OIyI7s.png)","metadata":{}},{"cell_type":"markdown","source":"Below I have visualised the starters success rate vs 1 Vise and 2 Vises. \n\nWhile many starters will regularly win in a one on one matchup, many will struggle to win when double teamed. This is a large opportunity for gunners as there are potentially ways to learn from successful gunners techniques which can be applied to their game to improve their own success rate.","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/qaTRj0u.png)","metadata":{}},{"cell_type":"markdown","source":"To add depth to the current data about the gunners I added their speed and acceleration metrics from the tracking data so we know how they compare against their peers. \n\nTo calculate the speed and acceleration metrics:\n1. On each play identified the max speed and acceleration within the first 3 seconds on play (limited to first 3 seconds as acceleration spikes if a player attempted a tackle)\n2. Took the median max speed and acceleration for each play as their final overall metric\n3. As gunners and vises are some of the fastest athletes in the world they are all fast, due to this I ranked each of the gunners against every other starting gunner to identify the speed edge needed to outrun the competition\n\nWhen put into a correlation heat map there is a strong correlation between success rate vs 1 vise and the top speed of the gunner. Speed is a very important factor in 1:1 matchups. \n\nHowever when looking at the success rate vs 2 vises there is a very low correlation for success with either Speed, Acceleration or Weight. This suggests there may be outside factors such as technique which may affect success more than just physical attributes.","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/uKeN0iq.png)","metadata":{}},{"cell_type":"markdown","source":"# 4. Breaking down and clustering gunner and vise techniques\n\nGunners have a very low success rate when matched up against 2 vises, because of this I wanted to deepdive into the techniques used by Gunners and Vises to see what the optimal tactical decisions are. \n\nTo achieve this I needed to cluster Gunner and Vise tracking data to identify what they were doing for each play, the technique used was K-means clustering on normalised data.\n\nClustering process:\n\n1. The first stage was cleaning the (x,y) tracking data into a consistent format suitable for clustering. This involved converting all gunner/ vise plays so that they were from the perspective of a gunner on the right hand side of the field with the (0,0) coordinate being the gunners position when the ball was snapped. A negative x value would be a location to the left of the gunner and closer to where the ball was snapped, a positive x value would be closer to the sideline. The (x,y) coordinates by frame were then transposed onto one line for each matchup to enable clustering.\n\n2. Once the data was normalised I started a two step clustering process. \n    - First clustering the pre snap location\n    - Second clustering on the post snap movement \n    \n3. Clustering on Pre Snap Location identified three unique clusters:\n    - The standard lineup, two stationary vises  2-3 yards off from the gunner (95.7% of snaps)\n    - Vises in motion (3.7% of snaps)\n    - Vises far away from gunner or gunner far behind the line of scrimmage (0.6% of snaps)\n\nDue to time constraints I only continued clustering on the main cluster 'The standard lineup'\n\n4. Clustering on Post Snap movement\nAs the movement post snap has a lot of variety and is very complex I individually clustered each position for the first second post snap to identify original tactic as the later movement is reactionary. \n\nThis delivered the follow cluster results for each position:\n\nGunner:\n- Inside Release (towards the snap)\n- Outside Release (towards the sideline)\n\nVise (both inner and outer):\n- Press Release (approaches the gunner to jam them)\n- Normal Release (neutral behaviour)\n- Bail Release (no movement forward and rapidly move backwards)\n\nInside Vise clustering sample:","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/ZRDLvov.png)","metadata":{}},{"cell_type":"markdown","source":"## 4.1 What tactic works well for gunners?\n\nGunners initially seem have an easy decision to make as they have one of two options:\n- Inside release (chosen 46% of the time)\n- Outside release (chosen 54% of the time)\n\nHowever there is a vastly different success rate depending on which release is chosen:\n- Inside releases are successful 23.5% of the time \n- Outside releases which are only successful 8% of the time.\n\nInside releases are almost x3 as successful and many gunners are not making the optimal choice. This makes sense from a game perspective as the sideline acts as another defender limiting the gunners movement. ","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/FLsNGvg.png)","metadata":{}},{"cell_type":"markdown","source":"## 4.2 What tactic works well for vises?\n\nVises however have three options to choose from:\n- Press release (chosen 29% of the time)\n- Normal release (chosen 51% of the time)\n- Bail release (chosen 20% of the time)\n\nThis means that between both vises there are 9 potential combinations of technique. These also then depend on decision made by the gunner of which release they choose.\n\nIf the gunner decides on an inner release the optimal vise technique:\n- Inside vise to have a press release\n- Outside vise to have a bail or normal release\n\nIf the gunner decides on an Outside release the optimal vise technique:\n- Inside vise to have a bail release\n- Outside vise slight preference for press release, but any release is very successful\n\nIn summary the best decision is depending on what way the gunner travels, the vise closest to the direction of travel to press the gunner and the vise furthest away to quickly drop. If the gunners in the league move to more inside releases the optimal default technique will be the inside vise in press release and outside vise in bail release.","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/6cJWWVA.png)","metadata":{}},{"cell_type":"markdown","source":"# 5. Overall tactical decisions\n\nIn the above sections we have covered tactical decisions at a player level, now we will compare the decisions at a team level. \n\nPunting teams are limited to two gunners by NFL rules, generally the punt return team either matches the gunners up \n- One to one both gunners\n- Double team one gunner, one to one the other\n- Double team both gunners\n\nFor punts you are likely wanting to maximise one of two possible outcomes: \n- Blocked punt to cause a turnover in great position to score from \n- Convert fair catches into returns to put your offense in a slightly better position\n\nHaving 2 vises in a one to one matchup leads to a 1.4% chance of a blocked or deflected punt which is almost double that of when you field 3 or 4 vises (0.4% and 0.8%).\n\nWhen looking at returns 72% of punts are returned where both gunners are double blocked vs, 32% for 2 vises and 57% for 3 vises.\n\nAs such during the game if you are needing to create a turnover in a hail mary situation 2 vises is the best team tactic, however if you are wanting a more reliable option to put your offense in a slightly better situation 4 vises is the optimal play.\n\n","metadata":{}},{"cell_type":"markdown","source":"![](https://i.imgur.com/3uruySa.png)","metadata":{}},{"cell_type":"markdown","source":"# 6. Summary & future work\n\nIn this analysis I have:\n1. Created a metric and framework for grading gunner performance\n    - In addition linked the metric to the result of the play\n\n\n2. Graded all starting gunners for the three seasons 2018 - 2020 and identifies key attributes for success\n    - Speed is the key ingredient for 1v1 matchups\n    - Technique is a key ingredient for 1v2 matchups\n\n\n3. Clustered gunner and vise behaviour pre and post snap to identify techniques used\n    - Identified opportunities to change behaviour to more successful techniques\n    - Identified optimal combinations of techniques for vises\n    \n    \n4. Identified the optimal punt formation for teams with a strong return game or creative special teams coach\n    - 2 vises in 1v1 matchup if desperately wanting to block a kick\n    - 4 vises, both doubling gunners if wanting to ensure a punt return \n    \n    \nWhile this analysis has identified several opportunities for improvement which can be quickly applied to the NFL it is far from an exhaustive overview of punting. \n\nFuture work could include researching the impact safeties have on the punt return and investigate the effectiveness of a punt returner and return tactics.\n\n\n\n\nThank you for reading this analysis, if you have any questions please reach out to me.\n\nCharles Giess","metadata":{}},{"cell_type":"markdown","source":"# 7. Appendix","metadata":{}},{"cell_type":"markdown","source":"Word count: 1982\nFigures used: 9","metadata":{}},{"cell_type":"markdown","source":"[Project Code](https://github.com/cejgiess/NFLBigDataBowl2022)","metadata":{}},{"cell_type":"markdown","source":"[LinkedIn](https://www.linkedin.com/in/charles-giess/)","metadata":{}},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}