Read Ebook {PDF EPUB} Cluster by Rider England Clustering riders: an improved approach. In this post we are going to continue with the clustering problem that we started in February. The idea remains the same, we are going to try to automatically group riders, but we changed our approach considerably. In the first step we identify four rider clusters: time trialists, sprinters, GC guys/climbers and classics specialists. After that we zoom in on the sprint cluster and the clustering algorithm comes up with three distinct sprinter types. The first sprinter type is the ‘high speed flat race’ type of guy, such as and . The second group of sprinters prefers long stages, and also scores in GC’s (Kuznetsov, Barbero, Mohorič). The third sprinter group consists is an all-round group that performs mainly on the World Tour level and like the classics as well. The most characteristic rider of this last group is Danny van Poppel and the top scorers here are e.g. Peter Sagan and . The old and new approach. In our first attempt we used the following dimensions to cluster riders in a single step: ODR flat Points scored in flat one day races ODR not-flat Points scored in one day races that are not flat ODR unknown Points scored in one day races for which we could not determine the profile TT Points scored in time trials during a race stage MT Points scored in mountain stages stage flat Points scored in flat stages stage hill Points scored in hilly stages stage hill-flat Points scored in hilly stages, but with a reasonably flat finish GC Points scored in GC’s (excluding youth/mountain/points/combat classifications) % WT Percentage of points scored at the UWT level % May Percentage of points scored before May 1. The result was 9 different clusters, but many clusters were quite alike and there were e.g. 3 clusters that were GC related, but the distinction came from the ‘% WT’ and ‘% May’ variables. These variables do not really characterize the type of rider but the level and moments of peak performance. Hence, it is more natural to look at these when they main groups are already determined. Furthermore, there is also a risk that riders from a different group (e.g. a sprinter) is put in a GC group because of great similarity in ‘% WT’ and ‘% May’. In this post we will take a different approach and do the following: We are going to make initial clusters on less dimensions. This most likely leads to fewer clusters that are clearly defined as GC, sprinter, etc. The eight dimensions that remain from the first version are: GC, ODR flat, ODR non-flat, TT, stage MT, stage flat, stage hill, and stage hill-flat. We introduce one new dimension: points from mountain classifications. Due to the fact that we count points for flat stages and hilly stages with a flat finish we do not award points for being in the final point classification. The total number of dimensions in the first step is nine. We stop using Zweeler points. Instead, we use a different point calculation where each top 20 classification yields points, with the number one earning 20 points and the number 20 gets 1 points. Riders get points for race stages, GC’s, MC’s (mountain classifications), and one day races. The GC / MC points are adjusted for race length, such that grand tours matter most. The 1.UWT classics points get a multiplier of 1.5. The points are re-scaled per rider as we are interested in the source of the points, not necessarily the number of points. This is unchanged compared to our first clustering post. Once we have obtained initial clusters, we are going to cluster again within each cluster to discover the different types of e.g. sprinters. In the second stage clustering procedure we will use additional variables such as ‘% WT’ and ‘% May’. An advantage of this approach is that we can change the second stage variables, dependent on the cluster that we investigate. For example, we will take a closer look at the sprinter cluster and use variables related to average race length and race speed. The first step clustering results are outlined in the next section. Step 1: general clusters. Figure 1 shows that, with our nine dimensions, the optimal number of clusters in the first step is four. The radial plot that shows the scores on each dimension for the four clusters is provided in Figure 2. Figure 1: Optimal number of clusters using silhouette appraoch. Figure 2: Scores per dimension for each cluster in the first step in a radial plot. From Figure 2 it quickly becomes clear that especially the classic specialists (cluster 4) and time trialists (cluster 1) gain their points from mainly one dimension. The sprinters in cluster 3 get their points from the different types of flat finishes, whereas the riders in cluster 2 score in mountain stages and GC’s. A bar chart representation of the scores on each dimension is provided in Figure 3. Figure 3: Scores per dimension for each cluster in the first step but now in a stacked bar chart. The four cluster outcomes are very logical and in fact these are the four categories on the PCS profiles (e.g. Alejandro Valverde)! It is a confirmation that the clustering method is sensible. In addition, it also shows that the profile scores on PCS are well thought of Before we will zoom in on the sprint cluster, we have outlined for each initial cluster the most characteristic riders (most exemplary) of each cluster as well as the riders with most points in each cluster. cluster 1: time trialists (15 riders) most characteristic: Jonathan Castroviejo, Nelson Oliveira, Maciej Bodnar, Simon Geschke, Stefan Küng top scoring: Jonathan Castroviejo, Nelson Oliveira, Maciej Bodnar, Simon Geschke, Stefan Küng. cluster 2: climbers / GC (185 riders) most characteristic: Pierre Latour, , , , Richard Carapaz top scoring: Pierre Latour, Sergio Henao, Tony Gallopin, Fabio Aru, Richard Carapaz. cluster 3: sprinters (97 riders) most characteristic: Danny van Poppel, Elia Viviani, Jempy Drucker, Clément Venturini, Peter Sagan top scoring: Danny van Poppel, Elia Viviani, Jempy Drucker, Clément Venturini, Peter Sagan. cluster 4: classics specialists (26 riders) most characteristic: , Robert Power, Sep Vanmarcke, Łukasz Wiśniowski, Michael Valgren top scoring: Oliver Naesen, Robert Power, Sep Vanmarcke, Łukasz Wiśniowski, Michael Valgren. To illustrate the concept of ‘most characteristic’ rider a bit more, take a look at Figure 4. Here you can see the scores on each dimension of the most characteristic sprinters in cluster 3. As you can see these riders indeed have quite a similar profile across all variables we consider in step 1. In the next section we will dive further into this sprint cluster and introduce new variables that allow us to split up this group further. Figure 4: Scores per dimension for first step sprint cluster 3. Step 2: re-clustering sprint cluster 3. For the second step we are going to focus on the 97 sprinters in cluster 3. We will cluster these sprinters again and use the following additional variables: % WT Percentage of points scored at the UWT level % May Percentage of points scored before May 1 team pos The average relative position of the rider within the team % high speed Percentage of points scored in the 25% of races with the highest average speed % 200+ km Percentage of points scored in races/stages with more than 200 kilometers. The additional variables should be reasonably self-explanatory, except for the ‘team pos’ variable. We introduce this variable to account for the fact that a lead-out may very well end up in the top 20 as well, but most likely behind the protected rider. Since we do not use the total number of points scored in the clustering we do look at how high ranked is a rider within his team. So for each race we look at the rank of a rider within his own team, re-scale this on the 0-100 domain and average it over all races the rider participates in. With the additional five variables (now 14 in total) the optimal number of clusters for the 97 sprinters is three (Figure 5) and the scores of the three clusters on each of the 14 dimensions can be found in Figure 6. Figure 5: Optimal number of clusters when re-clustering the sprinters from step 1 using the silhouette appraoch. Figure 6: Scores per dimension for each second-stage sprint cluster in a radial plot. The three sprint clusters show quite some variation in the new variables, in particular % WT and % high speed. It is not that straightforward to describe each cluster in just a few words, but we have tried our best. Cluster 3.1 can be characterized as all-round sprinters and classics riders at the world tour level, cluster 3.2 are the sprinters that score some GC points and love races of 200 kilometers or more. Finally cluster 3.3 is occupied by the pure strength, high-speed flat stage guys. The five most characteristic riders and riders with most points in each cluster are listed below. cluster 3.1: World Tour All-Round/Classics (30 riders) most characteristic: Danny van Poppel, Sacha Modolo, Mike Teunissen, Edward Theuns, Matteo Trentin top scoring: Peter Sagan, Elia Viviani, Greg Van Avermaet, Jasper Stuyven, Sonny Colbrelli. cluster 3.2: Long distance / GC (27 riders) most characteristic: Vyacheslav Kuznetsov, Carlos Barbero, Nils Politt, Magnus Cort, Matej Mohorič top scoring: Matej Mohorič, Carlos Barbero, Magnus Cort, Edvald Boasson Hagen, Patrick Bevin. cluster 3.3: High speed / super flat (40 riders) most characteristic: Giacomo Nizzolo, Dylan Groenewegen, Álvaro José Hodeg, Kristoffer Halvorsen, Luka Mezgec top scoring: Alexander Kristoff, Arnaud Démare, Dylan Groenewegen, Pascal Ackermann, Marc Sarreau. Finally, lets zoom in on the all-round sprint cluster 3.1. In Figure 7 we plotted the scores of the most characteristic riders in cluster 3.1, whereas in Figure 8 you can find the best scoring riders in this cluster. From the most characteristic riders in this cluster Sacha Modolo stands out on the hill- flat stages, and Mike Teunissen is in this cluster relatively strong on the GC dimension. When we look at the riders with most points we notice that Peter Sagan collects all his points from World Tour races. This is consistent with our extensive analysis of Sagan’s historic results. Figure 7: Scores per dimension for the most characteristic all round spinters in cluster 3.1. Figure 8: Scores per dimension for the best scoring all round spinters in cluster 3.1. Conclusion. The two-step approach when clustering riders is a considerable improvement over the single step approach that we tried earlier. The first step yields four clearly defined clusters consisting of time trialists, sprinters, GC guys/climbers and classics specialists. In the second step we looked at sprinters. Within the sprinter cluster we obtained three subgroups: flat stage specialists, sprinters with a GC component that like long distances and World Tour level all-round/classics sprinters. We will most likely follow up on this clustering post by investigating how these clusters members are distributed over the teams. If you have other suggestions don’t hesitate to send us a message on Twitter or email. BM Bikes & BM Riders Club. If you stop calling it "voltage" and call it potential difference (<< its proper name) then it is easier to grasp that you are measuring differences only, not absolute values. The intentional load is the bulb so ideally we would like the "potential difference" across the bulb to be the same as the battery voltage, that way all the energy is going into lighting the bulb up. In the real world that is impossible because any connections and wires introduce a small load of their own that we try to minimise. We can get an idea of how near to ideal it is by seeing how much potential difference we are getting between the bulb positive and the battery positive. Ideally zero but never quite getting there. Re: '87 R65 instrument cluster wiring. Post by r75boxer » Mon Aug 10, 2020 1:49 pm. Thanks once again Rob. I've testing the positive terminal on the battery (using the positive wire from the multimeter) and the ground on the bulb (brown wire) with the negative lead on the multimeter. That's what I thought you wanted (before I carefully read the instructions). Anyway, it's clear now and I will conduct the test as above. Re: '87 R65 instrument cluster wiring. Post by wulfrun » Mon Aug 10, 2020 2:37 pm. At the risk of your wrath Rob and excuse my ignorance, but surely the reading from the first bulb’s positive and battery positive would be 12 volts? Looking at the picture here on voltage drops: This is something I often struggle with; actually electrics in general. It's (nominally) 12V at the battery positive, but only assuming you're measuring to the battery negative. However, that's not the point of what Rob intended. At the bulb's positive terminal some voltage has been lost due to wiring/switches/contacts not being zero ohms. By measuring from the battery positive to the bulb positive, you find out what's being "lost" and whether it's acceptable or indicates a fault. Likewise measuring from battery negative to the bulb-holder negative can show an earth fault up. Since the voltage ought to be very low, you can use a more sensitive range on the meter and get a more accurate idea, once you've established it actually IS a low figure (i.e. no major fault). As an aside, the battery positive isn't even at (nominally) 12V, at least not unless you physically earth the frame to the ground (not practical if on the move!). It's just 12V more positive than the frame is. Re: '87 R65 instrument cluster wiring. Post by windmill john » Mon Aug 10, 2020 5:10 pm. Thanks all. I think I was reading it as the voltage at the battery positive to bulb positive, Rob was expecting 0.something voltage, not 12 less a bit of loss. I understand losses, just couldn’t follow the thread; my fault. Re: '87 R65 instrument cluster wiring. Post by Mjolinor » Mon Aug 10, 2020 5:23 pm. Thanks all. I think I was reading it as the voltage at the battery positive to bulb positive, Rob was expecting 0.something voltage, not 12 less a bit of loss. I understand losses, just couldn’t follow the thread; my fault. Now your confusion is showing. If you measure battery positive to bulb positive it is ideally zero but actually slightly above zero. The twelve volts that you have available from the battery is not of any consequence as you are not using the negative side at all, it could be running off a hundred volts and the result would be the same. Potential DIFFERENCE. Re: '87 R65 instrument cluster wiring. Post by windmill john » Mon Aug 10, 2020 7:15 pm. Ahh. so. making myself look even dafter possibly. would that explain the tiny figure Rob was talking about? A little acceptable loss? Or shall I just stick to carbs and mechanicals. Re: '87 R65 instrument cluster wiring. Post by r75boxer » Mon Aug 10, 2020 7:33 pm. Okay. After a few domestic errands I'm back at it. I believe the only question that needs to be anwered is #4. "4) The test readings should be taken at the positive bulb connection for each of the four indicator bulbs. TYhe wire colour will be Blue/Red or Blue/Black. The bulb in the circuit being tested should be on at the time. This means you'll have to do the left bulbs with the indicator switch to the left and the right bulbs with the switch to the right." With ignition on, left signal switch on (solid, not blinking) I get a reading of 0.004v on both bulbs. The right side, right signal on (solid) I get a reading of 0.002v on both bulbs. I repeated the measurement twice and got the same readings. Re: '87 R65 instrument cluster wiring. Post by Mjolinor » Mon Aug 10, 2020 8:11 pm. If you use the quote tags it makes reading the posts much much easier if you are quoting previous replies. As shown but remove the spaces: [ q u o t e ] Text in here [ / q u o t e ] What it looks like if you do it right. Re: '87 R65 instrument cluster wiring. Post by Rob Frankhamr » Mon Aug 10, 2020 8:40 pm. Okay. After a few domestic errands I'm back at it. I believe the only question that needs to be anwered is #4. "4) The test readings should be taken at the positive bulb connection for each of the four indicator bulbs. TYhe wire colour will be Blue/Red or Blue/Black. The bulb in the circuit being tested should be on at the time. This means you'll have to do the left bulbs with the indicator switch to the left and the right bulbs with the switch to the right." With ignition on, left signal switch on (solid, not blinking) I get a reading of 0.004v on both bulbs. The right side, right signal on (solid) I get a reading of 0.002v on both bulbs. I repeated the measurement twice and got the same readings. That sounds actually very good. With the knowledge that the lights are now working as they should and those readings , I would say it's time to stop worrying and ride the bike. Frankhams retirement home for elderly Boxers. Re: '87 R65 instrument cluster wiring. Post by Rob Frankhamr » Mon Aug 10, 2020 9:11 pm. Ahh. so. making myself look even dafter possibly. would that explain the tiny figure Rob was talking about? A little acceptable loss? Or shall I just stick to carbs and mechanicals. OK, you asked for it. Every material has a resistance (No I'm not going to get into superconductors . I'm talking in a world here that relates to practical motorcycle mechanics!). The resistance of any conductor is proportionate to it's length, the inverse of it's cross sectional area and a constant unique to the material it is made from. The constant is known as the 'coefficient of resistivity' for that material (I'm sure you really wanted to know that . Theres also another constant called the 'temperature coefficient of resistance' but, i the interests of KISS, forget I mentioned it ). The upshot of this is that every wire has a resistance value. If you have a current flowing through a resistance there will be a voltage drop across it. that is to say the voltage at the end connected closest to positive will alway be higher than the voltage at the end connected closest to negative. In a practical circuit, the lower the resistance in the wire, the lower the voltage across that resistance.. or to put it another way, there will always be a voltage between the ends of the wire. and that voltage is often referred to as a voltage drop (unintended voltage drop would be more accurate). The trick is to get it as low as possible. This can be done in a number of ways: a) keep the length of the wire as short as practical. the shorter the wire the less resistance. b) Use a thicker wire. the thicker the wire the less resistance. c) Reduce the number of high current loads attached to a single source. the higher the current, the greater the voltage drop. d) Reduce the number of conectors. each connector adds it's own resistance. and keep them as clean as possible. e) Use only good quality switches. Contacts also add a measure of resistance and, therefore, voltage drop. Clustering riders. Today we address a complex, but fun and interesting problem: clustering riders. This is by far a new idea, everybody knows Elia Viviani and Dylan Groenewegen are classified as ‘sprinters’, whereas Chris and Froome are ‘GC guys’. The label a rider receives is simply based on past results. We are going to investigate whether we can take this a step further by (1) using a uniform point system for race results and (2) including other dimensions than just outcomes. One of our findings is that although Dylan Groenewegen and Elia Viviani are both sprinters, they clearly belong to different clusters. Let’s dive in and investigate how we came to this conclusion. Truth to be told, clustering is an art as well as a science. It is a very exploratory type of analysis where you try to find subgroups of observations (in our case riders) that are most similar across the dimensions that you consider. Clustering requires substantial thought about the algorithm to use, the outcomes you include and whether you scale them in a particular manner. The results are in general not right or wrong, but should help you gain insights in a particular field. In our rider cluster procedure we consider the following 11 dimensions: ODR flat Points scored in flat one day races ODR not-flat Points scored in one day races that are not flat ODR unknown Points scored in one day races for which we could not determine the profile TT Points scored in time trials during a race stage MT Points scored in mountain stages stage flat Points scored in flat stages stage hill Points scored in hilly stages stage hill-flat Points scored in hilly stages, but with a reasonably flat finish GC Points scored in GC’s (excluding youth/mountain/points/combat classifications) % WT Percentage of points scored at the UWT level % May Percentage of points scored before May 1. All variables are calculated for the riders that are part of the 2019 World Tour teams. We require that a rider scores at least 50 points and participates in at least 20 single day races or race stages before we include him in the analysis. Nine out of 11 variables use ‘points’. We do not use UCI points or rank scores for the analysis. Instead, we use a point system very similar to the Zweeler games, where finishing first in a race stage yields 35 points. For a one day race a rider can earn up to 120 points. With this point system it pays off to finish in the top 20 (stage) or top 25 (one day race). It is good to know that we rescale the points variables per rider. For example, ‘TT’ is the percentage of the total rider points scored in time trials. The GC points are calculated as the number of stage points times the square root of the race length. As a consequence a victory yields the same amount of points as about 4.5 stage wins. The points may not be perfectly balanced, but keep in mind the main purpose of the points is to determine the strengths of riders, we will not really use them to claim one rider is ‘better’ than another. Unlike UCI points the Zweeler points to not change by race class (World Tour, HC, etc). We really do this on purpose, it allows us to identify riders that are similar to e.g. Peter Sagan or Greg van Avermaet in their race preferences and performance, but don’t show this at World Tour level. This could be because of their race preferences, perhaps they simply cannot compete with the absolute top or have team orders that prevent them from riding for their own chances. To compensate for the level the results were obtained on we have ‘% WT’ which is the percentage of points scored at the World Tour level. With the variable ‘% May’ (the percentage of points scored before May 1st) we try to capture the form peak of riders. Now before we can show some results we are left with one big question. How many groups of riders are there? Fortunately, this is not a trial and error process. There are several techniques that can be used to determine the number of clusters. One of these approaches is the ‘silhouette’ method and the results are shown in Figure 1. The method indicates the optimal number of groups is 9, however, the difference with 7 or 8 clusters is small. For now we stick with 9 clusters and let’s see what comes out. Figure 1: Optimal number of clusters using silhouette appraoch. Just to recap, we have 11 variables / dimensions and the algorithm indicates we should make 9 groups / clusters of riders with those variables. The resulting clusters are presented in two different ways. Figure 2 shows the scores of each cluster in a radial plot. It is quite intuitive to read and observe the differences between clusters. For example, lets focus on the grey-blue line (cluster number 1). The riders in this cluster score very high in the variables % WT and % May and take most those points from not flat single day races (ODR not-flat). Compare this with cluster 9. These riders also score high on the not flat single day races, but do this in general after May and not so much on a world tour level. Figure 2: Scores per dimension for each cluster in a radial plot. Some big names from cluster 1 are Greg van Avermaet and Peter Sagan, but guess who is in cluster 9: Alejandro Valverde . Well, yeah that can happen, if you let a computer do the job, but let’s just take a quick peek whether this makes sense given the variables. Figure 3 shows the scores for the most characteristic riders for cluster 1 (Daryl Impey) and cluster 9 (David Gaudu) as well as Peter Sagan (cluster 1) and Alejandro Valverde (cluster 9). The difference between the two groups can be explained by, among others, the percentage of points at World Tour level and points scored in hilly stages. Furthermore, Valverde and Gaudu tend to score more in GC’s. At some point we will also consider more advanced cluster types where each rider can actually be part of multiple clusters. It is likely that members of cluster 1 are also to some extent part of cluster 9 and vice versa. Figure 3: Scores per dimension for D. Impey and P. Sagan (cluster 1) / A. Valverde and D. Gaudu (cluster 9) If you are completely confused by these ‘spider webs’, take a look at a different representation of the clusters in Figure 4. The figure contains the scores of each cluster in a single bar. Each bar has colors that correspond to the variables and larger/smaller chunks indicate that the members of that cluster score in general high/low on that outcome measure. With these bars we made an attempt to characterize each cluster in a few words. These short descriptions are shown in Table 1. In this table you can also find the 5 most characteristic riders in each cluster. The characteristic riders should be exemplary for the cluster. Furthermore, the table also contains the riders that scored most points in each cluster. Please remember, this does not necessarily mean that these are the best riders in the cluster. Figure 4: Scores per dimension for each cluster in a stacked bar chart. Since this is already quite a lengthy post let’s check out the sprint clusters 3 and 7 characterized by Giacomo Nizzolo and Fabio Jakobsen, respectively. The radial/spider plots of these clusters are shown in Figure 5 and Figure 6. The differences between the two clusters are actually quite obvious. Cluster 3 contains sprinters that are really focused on the flat race stages, and get most of their points from World Tour races. The sprinters in cluster 7 still do very well on the flat stages (look at the points relative to the dashed blue circle), but score quite some points in one day races. Figure 5: Scores per dimension for sprint cluster 3. Figure 6: Scores per dimension for sprint cluster 7. Hopefully you enjoyed the insights. We are aware that there is plenty of room for improvement. In the future we will look more into the variables included in the analysis and we will also investigate whether it is insightful to allow for multiple cluster memberships. If you have any other suggestions, or if you would like to see some more figures of different clusters let us know on Twitter or by mail. 10 Surprising Facts About General Custer. 1. Four other members of the Custer family died at the Battle of Little Bighorn. Among the force of more than 200 men wiped out by the Lakota Sioux and Cheyenne warriors on June 25, 1876, were Custer’s 18-year-old nephew, Henry Reed, brother-in-law James Calhoun and two younger brothers, Boston and Thomas (a Civil War veteran and two-time Medal of Honor recipient). 2. His nickname was “Autie.” Custer’s mispronunciation of his middle name when he first began to speak was adopted by his family as his nickname. The moniker stuck with him for his entire life and was used by his wife, Libbie, as a term of endearment. 3. Custer graduated last in his class at West Point. Custer was known by his fellow cadets at the U.S. Military Academy as the “dare-devil of the class” who devoted more energy to pranks than to his academic studies. Custer’s voluminous record of demerits earned him extra guard duty on most Saturdays, but he did manage to graduate from West Point in 1861, albeit as the lowest-ranking cadet, now known as “the goat.” 4. Custer became a Civil War general in the Union Army at 23. Although Custer struggled in the classroom, he excelled on the battlefield. After joining the Army of the Potomac’s cavalry following his graduation, he gained notice for his daring cavalry charges, bold leadership style and tactical brilliance. In June 1863, Custer was promoted to the rank of brigadier general at the age of 23, and he cemented his reputation as the “Boy General” days later at the Battle of Gettysburg when he repelled a pivotal Confederate assault led by J.E.B. Stuart. By the end of the Civil War, Custer had risen to the rank of major general. (The youngest Civil War general is believed to have been Galusha Pennypacker, who was 20 when he became a brigadier general for the Union Army.) 5. His wife received the table on which the Civil War surrender terms were drafted. Custer served the Army of the Potomac with distinction from the First Battle of Bull Run in 1861 to Appomattox Court House four years later. Custer’s forces blocked Confederate General Robert E. Lee’s final retreat, and he received the white truce flag signifying Lee’s wish to meet with Union General Ulysses S. Grant. Custer was present in the front parlor of Wilmer McLean’s home on April 9, 1865, when Lee surrendered to Grant. General Philip Sheridan purchased the small spool-turned table used by Grant to draft and sign the terms of surrender and gave it to Libbie Custer in gratitude for her husband’s service. “Permit me to say, Madam,” Sheridan wrote to Libbie, “that there is scarcely an individual in our service who has contributed more to bring about this desirable result than your gallant husband.” After Libbie’s death in 1933, her will specified that the table be given to the Smithsonian Institution. 6. Custer scented his hair with cinnamon oil. The flamboyant Custer paid great attention to his appearance. He wore a black velvet uniform with coils of gold lace, spurs on his boots, a red scarf around his neck and a large, broad-brimmed sombrero. Custer took particular pride in his cascading golden locks, which he perfumed with cinnamon oil. 7. Custer was twice court-martialed. Prior to graduating from West Point, Custer was court-martialed for neglect of duty in failing to stop a fight between two cadets when he was officer of the guard and received a light punishment. Much more serious was his 1867 court-martial in which he was convicted on eight counts that included conduct to the prejudice of good order and military discipline, and absence without leave from his command after he left part of his regiment on the Kansas frontier while he returned without orders to Fort Riley in order to see Libbie. Custer was suspended from rank and command without pay for a year, but Sheridan reinstated him after 10 months to lead a bloody campaign against the Cheyenne. 8. Buffalo Bill helped to mythologize Custer’s Last Stand. After Custer’s death, his widow devoted the remaining 57 years of her life writing books and delivering lectures that cast her husband as a martyr who gallantly staged his “Last Stand.” William “Buffalo Bill” Cody, who had briefly scouted for Custer, also contributed to the mythmaking. Weeks after the Battle of Little Bighorn, he killed and scalped a Cheyenne warrior named Yellow Hair and declared it “the first scalp for Custer.” Buffalo Bill replayed the scene repeatedly throughout his theatrical career and incorporated a re-enactment of “Custer’s Last Rally,” complete with several Native Americans who had actually been present at the Battle of Little Bighorn, into his famed Wild West show. 9. Custer was thought to have lived a charmed life. During the Civil War, the “Boy General” seemed to have such a streak of good fortune, which included his avoidance of serious injury in spite of his daring command and having 11 horses shot out from under him, that is was referred to as “Custer’s luck.” The revival of his military career after his 1867 court-martial furthered the perception that Custer lived a charmed life, but Custer’s luck ran out at Little Bighorn. What is it to Cluster? Founded in 2016, Cluster set the tone for art fairs that do away with middle men, dropped gallery representation and percentages skimmed off the top. Instead, it set out to be London’s most affordable art fair, featuring pure creation, unmitigated and unapologetic. Since then Cluster has grown into a celebration of diversity, style, material and themes, of ethnicity, background, age and class, a melting pot defined solely by the artistic necessity to express. To date Cluster has held eleven fairs, including ones dedicated to Illustration, Craft, Jewellery and Photography. With over 300 creatives featured, Cluster has formed a community of practitioners, which it supports through a programme of monthly gatherings, career support, one-on-one development sessions, talks, seminars and workshops, as well as the opportunity to take part in an ongoing series of curated projects and more. Cluster is more than a fair, it is a joint venture, a share journey, creative safe house and exercise in community building. Each fair and every event is a meeting place of like-minded individuals, sharing stories of success and struggle, initiating conversations and collaborations.