31:T458,list = (() => { const items = Array.from(itemsWithOccurrence).map(([itemId, occurence]) => { const item = itemData.get(itemId) if (!item || item.types.includes("gun")) { return null; } const totalPrice = item.lastLowPrice const pricePerSlot = Math.round(totalPrice / (item.width * item.height)) return { icon: item.iconLink, name: item.shortName, pricePerSlot, occurence, name: shortNames.get(itemId) } }).filter((item) => item && item.pricePerSlot > minPricePerSlot).sort((a, b) => b.occurence - a.occurence).slice(0, numberOfItemsToShow2) const chunks = splitIntoChunks(items, numberOfChunks).map((chunk) => chunk.sort((a, b) => b.pricePerSlot - a.pricePerSlot)) return htl.html`

${chunks.map((chunk) => htl.html.fragment`

${chunk.map((item) => htl.html.fragment`

`)}

` })()32:T495,listWithMedianPrices = (() => { const items = Array.from(itemsWithOccurrence).map(([itemId, occurence]) => { const item = itemData.get(itemId) if (!item || item.types.includes("gun")) { return null; } const medianPrice = d3.median(item.historicalPrices.map((price) => price.price)) const pricePerSlot = Math.round(medianPrice / (item.width * item.height)) return { icon: item.iconLink, name: item.shortName, pricePerSlot, occurence, name: shortNames.get(itemId) } }).filter((item) => item && item.pricePerSlot > minPricePerSlot).sort((a, b) => b.occurence - a.occurence).slice(0, numberOfItemsToShow2) const chunks = splitIntoChunks(items, numberOfChunks).map((chunk) => chunk.sort((a, b) => b.pricePerSlot - a.pricePerSlot)) return htl.html`

${chunks.map((chunk) => htl.html.fragment`

${chunk.map((item) => htl.html.fragment`

`)}

` })()33:[{"id":"e50958a50696651967fecd00e55cc14ee4bd8498f8b7e85190747e6fd0ad01fa15102313a8f1dbca32170bbbc6bfd884f97e76bc16c318b8ca18425f4201ba8e","href":"https://static.observableusercontent.com/files/e50958a50696651967fecd00e55cc14ee4bd8498f8b7e85190747e6fd0ad01fa15102313a8f1dbca32170bbbc6bfd884f97e76bc16c318b8ca18425f4201ba8e","name":"staticContainers.json","type":"application/json","lastModified":1706951207539.683,"size":2223343},{"id":"250c1ed685b9bb87d89494480a6caf5d62e38cd0b2cba8b791710aadee13a70576b943b4b8d6d33a91ada725dd984c6e8784276d072936d994524e58aad98586","href":"https://static.observableusercontent.com/files/250c1ed685b9bb87d89494480a6caf5d62e38cd0b2cba8b791710aadee13a70576b943b4b8d6d33a91ada725dd984c6e8784276d072936d994524e58aad98586","name":"staticLoot.json","type":"application/json","lastModified":1706951201082.174,"size":613168},{"id":"bb9d15c741834349578c44488e90d457b279d1494411c62bae225c44231392890d7e06a4d08d4edac3d5c6027237c4e3518e87f13bf580451192c2b9bf35ff60","href":"https://static.observableusercontent.com/files/bb9d15c741834349578c44488e90d457b279d1494411c62bae225c44231392890d7e06a4d08d4edac3d5c6027237c4e3518e87f13bf580451192c2b9bf35ff60","name":"en.json","type":"application/json","lastModified":1706951229763.028,"size":1880244}] 2f:{"title":"Tarkov Loot Experiments","cells":[{"id":0,"value":"# Tarkov Loot Experiments","pinned":false,"hidden":false,"mode":"md"},{"id":311,"value":"Note that the container and item distribution data used here is gathered in offline raids. It's possible that online raids have different configurations, and likely that the data is at least somewhat out of date. That being said, I think the data is still good enough to be useful.","pinned":false,"hidden":false,"mode":"md"},{"id":346,"value":"My goal is to create a tool to help optimize loot runs in terms of per-slot value. Many item tier lists exist for this purpose, but they usually only take into account item value as their means to sort items, and aren't updated automatically (because they are user-curated, or statically generated into a PDF or image). In order to be the most useful, I want to also take into account the rarity of items by presenting common but still worthwhile items first, because those are the ones users are most likely to want to check quickly. ","pinned":false,"hidden":false,"mode":"md"},{"id":367,"value":"This is some experimentation towards that project idea. I really wanted to try out observable, so this seemed like a good fit.","pinned":false,"hidden":false,"mode":"md"},{"id":458,"value":"## Estimating Rarity","pinned":false,"hidden":false,"mode":"md"},{"id":315,"value":"The central part of this experimentation is estimating the rarity of items. To do this, we first get the average number of containers of a particular type, per map. Since we already have a probability for each location the container can spawn in on a given map, we can just sum them to get the average number.","pinned":false,"hidden":false,"mode":"md"},{"id":27,"value":"staticContainersPerMap = new Map(\n Object.keys(containers).map((mapName) => [\n mapName,\n containers[mapName].staticContainers.reduce((containerCounts, currentContainer) => {\n // reduce id to name here because there are multiple containers with the same name but different ids (in particular\n // noticeable with different weapon crates)\n // making them distinguishable seems like lots of work for little benefit so we just combine them into one early\n const containerName = shortNames.get(currentContainer.template.Items[0]._tpl)\n return containerCounts.set(containerName, (containerCounts.get(containerName) ?? 0) + currentContainer.probability)\n }, new Map())\n ])\n)","pinned":false,"hidden":false,"mode":"ojs"},{"id":221,"value":"viewof mapsToShow = Inputs.checkbox(staticContainersPerMap.keys(), { label: \"maps to show\", value: staticContainersPerMap.keys() })","pinned":false,"hidden":false,"mode":"ojs"},{"id":136,"value":"Plot.plot({\n title: \"Average Number of Containers per Map\",\n marginRight: 90,\n marginBottom: 100,\n width: width,\n x: {\n tickRotate: 30,\n grid: true\n },\n marks: [\n Plot.frame(),\n Plot.barY(\n Array.from(staticContainersPerMap).filter(([mapName]) => mapsToShow.includes(mapName)).flatMap(([mapName, containers]) => (\n Array.from(containers).map(([containerName, count]) => ({ containerName, count, mapName }))\n )), {\n x: \"containerName\",\n y: \"count\",\n fy: \"mapName\"\n }\n )\n ]\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":319,"value":"Knowing the average number of containers, we want to find out what's in them. The original data has a field `relativeProbability` for every item per container, which I interpret to mean that when dividing it by the sum of all of these \"relative probabilities\" we get an actual probability. Also, containers can have more than one item in them, so we multiply by the average number of items for the given container type. This isn't perfect because some items are too big to fit multiple times in a particular container, but it doesn't look like it skews the data too much (we would expect to see bigger items dis-proportionally often, which doesn't seem to be the case to me).","pinned":false,"hidden":false,"mode":"md"},{"id":58,"value":"itemsPerContainer = Object.keys(loot).reduce((itemsPerContainer, currentContainerTpl) => {\n const currentContainer = loot[currentContainerTpl];\n const averageItemCount = normalizeProbabilities(currentContainer.itemcountDistribution)\n .reduce((averageCount, current) => averageCount += current.count * current.probability, 0)\n\n // see above for early name conversion reasoning\n const containerName = shortNames.get(currentContainerTpl)\n const items = itemsPerContainer.get(containerName) ?? new Map()\n \n const normalizedItemDistribution = normalizeProbabilities(currentContainer.itemDistribution)\n for (const item of normalizedItemDistribution) {\n // ...and convert here early as well for consistency\n items.set(item.tpl, (items.get(item.tpl) ?? 0) + item.probability * averageItemCount)\n }\n \n return itemsPerContainer.set(containerName, items);\n}, new Map())","pinned":false,"hidden":false,"mode":"ojs"},{"id":214,"value":"viewof containersToShow = Inputs.checkbox(itemsPerContainer.keys(), {\n label: \"containers to show\", value: [Array.from(itemsPerContainer.keys())[0]]\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":189,"value":"htl.html`

${Array.from(itemsPerContainer).filter(([containerName]) => containersToShow.includes(containerName)).map(([containerName, items]) =>\n Plot.plot({\n width: width,\n title: containerName,\n marginBottom: 50,\n x: {\n tickRotate: 30,\n grid: true\n },\n marks: [\n Plot.frame(),\n Plot.barY(\n Array.from(items).map(([itemTpl, count]) => ({ itemName: shortNames.get(itemTpl), count }))\n .sort((a, b) => b.count - a.count).slice(0, 40),\n {\n x: \"itemName\",\n y: \"count\",\n sort: { x: \"y\" }\n }\n )\n ]\n })\n)}

`","pinned":false,"hidden":false,"mode":"ojs"},{"id":325,"value":"We now have all the data we need to estimate how common an item is. And, considering the graphs I generated from the data, it looks much like what one would expect. Since the volume of data is much too high to properly verify, this is a good sanity check.","pinned":false,"hidden":false,"mode":"md"},{"id":328,"value":"Unifying these two results into a single list of items sorted by probability is relatively trivial now, only made slightly annoying because I wanted to use `Map`s for everything.","pinned":false,"hidden":false,"mode":"md"},{"id":240,"value":"totalStaticContainers = Array.from(staticContainersPerMap.values()).reduce((totalStaticContainers, currentMapContainers) => {\n for (const [container, count] of currentMapContainers) {\n totalStaticContainers.set(container, (totalStaticContainers.get(container) ?? 0) + count)\n }\n return totalStaticContainers;\n}, new Map())","pinned":false,"hidden":false,"mode":"ojs"},{"id":242,"value":"itemsWithOccurrence = Array.from(itemsPerContainer).reduce((itemsWithOccurrence, [container, items]) => {\n for (const [itemName, itemProbability] of items) {\n itemsWithOccurrence.set(itemName, (itemsWithOccurrence.get(itemName) ?? 0) + totalStaticContainers.get(container) * itemProbability)\n }\n return itemsWithOccurrence\n}, new Map())","pinned":false,"hidden":false,"mode":"ojs"},{"id":353,"value":"viewof numberOfItemsToShow1 = Inputs.number({ label: \"number of items to show\", value: 40 })","pinned":false,"hidden":false,"mode":"ojs"},{"id":332,"value":"Plot.plot({\n width: width,\n marginBottom: 50,\n x: {\n tickRotate: 30,\n grid: true\n },\n marks: [\n Plot.frame(),\n Plot.barY(Array.from(itemsWithOccurrence).map(([itemTpl, occurrence]) => ({ item: shortNames.get(itemTpl), occurrence })).sort((a, b) => b.occurrence - a.occurrence).slice(0, numberOfItemsToShow1), {\n x: \"item\",\n y: \"occurrence\",\n sort: { x: \"y\" }\n })\n ]\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":344,"value":"Going by the graph, the data looks reasonable. But it is not very useful right now, we don't care about RUB as an item and many of the other top items are worthless. We need to take into account the item values.","pinned":false,"hidden":false,"mode":"md"},{"id":236,"value":"## First Tier List Draft","pinned":false,"hidden":false,"mode":"md"},{"id":363,"value":"Getting item prices is straightforward enough using an API, see the Appendix for the exact call and source. A first draft of a tier list that considers both price and rarity could look like this:\n\n1. sort all items by rarity, as we have done in the graph above\n2. get the per-slot price for all items\n3. discard items with a price below some threshold\n4. sort items by their rarity and split them into chunks\n5. sort items inside chunks by their value\n\nThis way we group items by their rarity, and establish a value hierarchy in the given groups.","pinned":false,"hidden":false,"mode":"md"},{"id":254,"value":"viewof minPricePerSlot = Inputs.number({ label: \"min price per slot in rub\", value: 10000 })","pinned":false,"hidden":false,"mode":"ojs"},{"id":301,"value":"viewof numberOfItemsToShow2 = Inputs.number({ label: \"number of items to show\", value: 200 })","pinned":false,"hidden":false,"mode":"ojs"},{"id":308,"value":"viewof numberOfChunks = Inputs.number({ label: \"number of chunks\", value: 4 })","pinned":false,"hidden":false,"mode":"ojs"},{"id":244,"value":"$31","pinned":false,"hidden":false,"mode":"ojs"},{"id":395,"value":"### Median Item Price","pinned":false,"hidden":false,"mode":"md"},{"id":422,"value":"The two-dimensional tier list seems to work pretty well with capturing what I want, but the prices on some items seem off. (In particular, at the time of writing, things like CPU fans were pretty far down in their chunk, even though they sell for a good bit. Also, things like Augmentin were not even on the list.) I think this is because I was using the last low price, which I chose to be on the safe side (underpromise, overdeliver). Instead of that, this iteration uses the median price of the item in the timespan that can be queried through the API (one week for most items, as far as I can tell).","pinned":false,"hidden":false,"mode":"md"},{"id":398,"value":"$32","pinned":false,"hidden":false,"mode":"ojs"},{"id":462,"value":"This looks much better, but naturally both options are useful.\n\nI will leave it at that as a proof of concept, if I ever get around to building out the full project that I am imagining I will link it here.","pinned":false,"hidden":false,"mode":"md"},{"id":21,"value":"## Appendix","pinned":false,"hidden":false,"mode":"md"},{"id":42,"value":"Loot, container data and translations from [SPT-AKI](https://dev.sp-tarkov.com/SPT-AKI/Server). (At the time of writing, the used files were in `project/assets/database/loot`.)","pinned":false,"hidden":false,"mode":"md"},{"id":7,"value":"translations = await FileAttachment(\"en.json\").json();","pinned":true,"hidden":false,"mode":"ojs"},{"id":100,"value":"shortNames = Object.keys(translations).reduce((keyToShortName, currentKey) => {\n const [tpl, type] = currentKey.split(\" \");\n if (type === \"ShortName\") {\n return keyToShortName.set(tpl, translations[currentKey])\n } else {\n return keyToShortName;\n }\n}, new Map())","pinned":false,"hidden":false,"mode":"ojs"},{"id":23,"value":"containers = await FileAttachment(\"staticContainers.json\").json()","pinned":true,"hidden":false,"mode":"ojs"},{"id":25,"value":"loot = await FileAttachment(\"staticLoot.json\").json()","pinned":true,"hidden":false,"mode":"ojs"},{"id":65,"value":"normalizeProbabilities = (array) => {\n const relativeProbabilitiesSum = array.reduce((acc, cur) => acc += cur.relativeProbability, 0)\n return array.map((item) => ({ ...item, probability: item.relativeProbability / relativeProbabilitiesSum }))\n}","pinned":true,"hidden":false,"mode":"ojs"},{"id":225,"value":"Price data from [tarkov.dev](https://tarkov.dev/api/).","pinned":false,"hidden":false,"mode":"md"},{"id":227,"value":"itemData = new Map((await (await fetch(\"https://api.tarkov.dev/graphql\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n query: `{\n items {\n id\n shortName\n iconLink\n width\n height\n types\n lastLowPrice\n historicalPrices {\n price\n timestamp\n }\n }\n }`\n })\n})).json()).data.items.map((item) => [item.id, { ...item }]))","pinned":true,"hidden":false,"mode":"ojs"},{"id":304,"value":"splitIntoChunks = (array, n) => {\n const copiedArray = [...array];\n \n const result = [];\n for (let i = n; i > 0; i--) {\n result.push(copiedArray.splice(0, Math.ceil(copiedArray.length / i)));\n }\n return result;\n}","pinned":true,"hidden":false,"mode":"ojs"}],"files":"$W33","stdlib":"1","titling":"auto"}