1 00:00:00,000 --> 00:00:07,880 Welcome to the Cannabis Data Science Meetup Group. 2 00:00:07,880 --> 00:00:10,480 As always, you're in for a treat today. 3 00:00:10,480 --> 00:00:14,600 We're going to crunch a lot of cool new cannabis data. 4 00:00:14,600 --> 00:00:20,380 Finally got our paws on the Washington State data for 2022. 5 00:00:20,380 --> 00:00:29,960 This is spanning, I believe, January 22nd or so of 2021 through the end of November, 6 00:00:29,960 --> 00:00:33,200 the beginning of December of 2022. 7 00:00:33,200 --> 00:00:36,600 We have almost a year worth of lab results. 8 00:00:36,600 --> 00:00:38,440 I'll share these with you. 9 00:00:38,440 --> 00:00:40,240 I've been curating them. 10 00:00:40,240 --> 00:00:48,680 There's maybe between 48 and 52,000 lab tests, lab samples. 11 00:00:48,680 --> 00:00:52,760 That's almost 2 million analytes that were tested. 12 00:00:52,760 --> 00:01:00,720 Really exciting, one of the largest sets of lab results, and it's in complete population, 13 00:01:00,720 --> 00:01:03,200 which makes it real fruitful for analysis. 14 00:01:03,200 --> 00:01:10,480 Instead of having a potentially non-random sample, we have each and every lab test that 15 00:01:10,480 --> 00:01:15,280 was conducted in Washington State, which gives us a lot of statistical power. 16 00:01:15,280 --> 00:01:20,680 Without further ado, I'll go ahead and just give you a brief presentation of what I'm 17 00:01:20,680 --> 00:01:23,320 interested in researching. 18 00:01:23,320 --> 00:01:29,440 As we may know, survey data may not be the most reliable data in the world because it 19 00:01:29,440 --> 00:01:32,960 relies on people to self-report. 20 00:01:32,960 --> 00:01:41,240 There's a lot of literature in psychology and in economics that suggests that people 21 00:01:41,240 --> 00:01:47,960 may not necessarily have the incentive to be honest with what they tell you. 22 00:01:47,960 --> 00:01:54,040 What they teach you in economics is look at people's actions rather than what they may 23 00:01:54,040 --> 00:01:55,040 say. 24 00:01:55,040 --> 00:01:58,320 People's actions hopefully will reveal their preferences. 25 00:01:58,320 --> 00:02:04,960 In this case, we're interested in many things. 26 00:02:04,960 --> 00:02:08,880 One of the things we started talking about in the past few weeks were contaminants. 27 00:02:08,880 --> 00:02:11,560 We started looking at microbes. 28 00:02:11,560 --> 00:02:15,600 In the past, we've looked at residual solvents. 29 00:02:15,600 --> 00:02:23,920 Now that we have 2022 data, pesticide testing was mandated in Washington State in April 30 00:02:23,920 --> 00:02:25,600 of 2022. 31 00:02:25,600 --> 00:02:33,680 Before that, other states have pesticide testing mandated, California, Oklahoma, perhaps I 32 00:02:33,680 --> 00:02:36,120 want to say Massachusetts. 33 00:02:36,120 --> 00:02:40,920 Most states already had pesticide testing mandated, but as we know, we don't have the 34 00:02:40,920 --> 00:02:44,120 best access to data in these states. 35 00:02:44,120 --> 00:02:51,680 For example, in California, the best we know of what pesticides are cultivators using is 36 00:02:51,680 --> 00:02:53,680 this self-reported survey. 37 00:02:53,680 --> 00:03:01,200 As we can see here, there's only 59 growers, so this leaves us wanting for more information. 38 00:03:01,200 --> 00:03:03,720 This was four years ago. 39 00:03:03,720 --> 00:03:05,120 It's been some time now. 40 00:03:05,120 --> 00:03:08,000 We've got some new data and some new techniques. 41 00:03:08,000 --> 00:03:13,720 Let's see if we can't replicate this per se, but let's see if we can't come close to 42 00:03:13,720 --> 00:03:19,880 replicating an analysis of pesticides being used in cannabis. 43 00:03:19,880 --> 00:03:25,320 As always, at the Cannabis Data Science Meetup Group, we're here to get our hands on the 44 00:03:25,320 --> 00:03:26,320 actual data. 45 00:03:26,320 --> 00:03:33,200 As I said, I was curating these data points for a cannabis data project. 46 00:03:33,200 --> 00:03:37,280 This is open source, and you can get your hands on this too. 47 00:03:37,280 --> 00:03:43,960 There's one thing that I want to iron out, and then I'll get this sent to you. 48 00:03:43,960 --> 00:03:52,520 Essentially, what I want to iron out, when I first did the first pass, my first pass 49 00:03:52,520 --> 00:03:57,360 just curated data from January to the end of November. 50 00:03:57,360 --> 00:04:04,800 It only curated around 48,000 unique inventory items that were tested. 51 00:04:04,800 --> 00:04:13,360 In the second pass, I collected all the lab results from January to December 7th, and 52 00:04:13,360 --> 00:04:18,240 that puts us around 52,000 lab results. 53 00:04:18,240 --> 00:04:19,920 It's not 100%. 54 00:04:19,920 --> 00:04:27,220 The samples don't overlap perfectly, so I want to rerun this and get a better merge 55 00:04:27,220 --> 00:04:29,080 of this data for you. 56 00:04:29,080 --> 00:04:38,040 However, right now, we do have more than 48,000 inventory items that were tested, so this 57 00:04:38,040 --> 00:04:43,640 can get us a good start at the very least. 58 00:04:43,640 --> 00:04:46,800 First things first, let's look at this data. 59 00:04:46,800 --> 00:04:50,500 What do we have here? 60 00:04:50,500 --> 00:04:57,480 These are the different types of products that were tested in Washington State in 2022. 61 00:04:57,480 --> 00:05:09,440 As we can see, flower lots were the bulk of testing, with just shy of 25,000 lots of cannabis 62 00:05:09,440 --> 00:05:11,080 flower being tested. 63 00:05:11,080 --> 00:05:13,900 Second, hydrocarbon concentrates. 64 00:05:13,900 --> 00:05:20,480 This appears to be the most popular type of concentrates, with more than 6,000 inventory 65 00:05:20,480 --> 00:05:22,680 items tested. 66 00:05:22,680 --> 00:05:31,520 Surprisingly, edibles were higher on this list than I anticipated. 67 00:05:31,520 --> 00:05:35,840 As you can see, there's different types of products for sure. 68 00:05:35,840 --> 00:05:43,880 However, for today, I thought maybe we could just look at the bulk, the flower lots, and 69 00:05:43,880 --> 00:05:50,800 the hydrocarbon concentrates, just to pick two categories, just to keep things relatively 70 00:05:50,800 --> 00:05:52,300 simple. 71 00:05:52,300 --> 00:05:56,960 This can be generalized to more sample types. 72 00:05:56,960 --> 00:05:58,640 This is a nice starting point. 73 00:05:58,640 --> 00:06:00,760 What are we after? 74 00:06:00,760 --> 00:06:03,440 What does this data even look like? 75 00:06:03,440 --> 00:06:12,440 I'll just show you a random sample here, just to show you all the cool data points we have. 76 00:06:12,440 --> 00:06:14,440 There's a lot of data for each of these. 77 00:06:14,440 --> 00:06:17,480 I'll just show you some of the cool things that we know. 78 00:06:17,480 --> 00:06:21,420 Also, I may have mislabeled this as the retailer. 79 00:06:21,420 --> 00:06:24,680 This may actually be the producer. 80 00:06:24,680 --> 00:06:27,500 I want to go through and double-check this. 81 00:06:27,500 --> 00:06:29,560 We know the name. 82 00:06:29,560 --> 00:06:33,600 This was cherry sorbet. 83 00:06:33,600 --> 00:06:35,200 We know the inventory type. 84 00:06:35,200 --> 00:06:37,680 This is cherry sorbet flower. 85 00:06:37,680 --> 00:06:42,160 This one coincidentally failed lab testing. 86 00:06:42,160 --> 00:06:46,520 We've got the results here. 87 00:06:46,520 --> 00:06:53,680 We should be able to dig in to the results to see why it failed testing. 88 00:06:53,680 --> 00:07:02,280 This one may have failed for microbes because there were no pesticides or residual solvents 89 00:07:02,280 --> 00:07:03,280 detected. 90 00:07:03,280 --> 00:07:08,160 Long story short, we've got some nicely formatted data. 91 00:07:08,160 --> 00:07:10,440 We'll just keep it visual. 92 00:07:10,440 --> 00:07:16,560 I'll just show you real quick the various pesticides that were detected in cannabis 93 00:07:16,560 --> 00:07:20,960 flower in Washington state in 2022. 94 00:07:20,960 --> 00:07:29,800 As you can see, piperanol butoxide takes the lead with more than 600 detections. 95 00:07:29,800 --> 00:07:34,240 Pesticides are notoriously hard to pronounce. 96 00:07:34,240 --> 00:07:44,240 As I mentioned in prior weeks, peculiar names, there may be more magic to the show than one 97 00:07:44,240 --> 00:07:46,200 may think. 98 00:07:46,200 --> 00:07:54,600 For example, I've heard a chemist make the point that one of the reasons why these chemicals 99 00:07:54,600 --> 00:08:02,320 may have odd names are for comprehension and memory retention. 100 00:08:02,320 --> 00:08:12,520 It's going to be hard for you to recognize, say, and remember imidacloripid. 101 00:08:12,520 --> 00:08:16,160 I probably mispronounced that. 102 00:08:16,160 --> 00:08:29,160 Just to show you a bit about these pesticides here, I thought I had one of them pulled up. 103 00:08:29,160 --> 00:08:33,400 Here's piperanol butoxide. 104 00:08:33,400 --> 00:08:37,400 This was what I was searching for in the prior week. 105 00:08:37,400 --> 00:08:40,400 This I think is, what's the name of this? 106 00:08:40,400 --> 00:08:43,080 A molecular structure. 107 00:08:43,080 --> 00:08:47,900 This is a way that you can depict chemical compounds. 108 00:08:47,900 --> 00:08:56,440 As I was telling you, I'm doing my best to read up on organic chemistry and probably 109 00:08:56,440 --> 00:09:04,360 know less than your average high school chemistry student. 110 00:09:04,360 --> 00:09:10,080 From what I'm gathering, this is just a simplified diagram. 111 00:09:10,080 --> 00:09:12,440 This is organic chemistry. 112 00:09:12,440 --> 00:09:18,700 The fundamental building block of organic chemistry is carbon. 113 00:09:18,700 --> 00:09:22,260 This is what life forms are made out of. 114 00:09:22,260 --> 00:09:30,800 It's real interesting how life works because you just have these fundamental building blocks. 115 00:09:30,800 --> 00:09:41,280 You've got carbon, which are essentially depicted from my understanding by these various lines. 116 00:09:41,280 --> 00:09:47,360 I'm going to butcher the interpretation of this, but from my understanding, these lines 117 00:09:47,360 --> 00:09:53,800 represent carbon bonding to oxygen. 118 00:09:53,800 --> 00:10:02,640 Instead of having carbon bonding to hydrogen or oxygen. 119 00:10:02,640 --> 00:10:04,640 Like I said, I'm butchering this. 120 00:10:04,640 --> 00:10:13,080 I'm not going to be able to think of this on the fly, but here you can see the white 121 00:10:13,080 --> 00:10:17,480 molecules, I forget if it's hydrogen or oxygen, aren't depicted. 122 00:10:17,480 --> 00:10:20,600 It must be hydrogen, I think. 123 00:10:20,600 --> 00:10:26,440 Yes, because see here at the end, this is H3. 124 00:10:26,440 --> 00:10:30,080 So three white balls and one gray ball. 125 00:10:30,080 --> 00:10:33,600 So I think that's three hydrogens and a carbon. 126 00:10:33,600 --> 00:10:37,840 And then the red must be the oxygen. 127 00:10:37,840 --> 00:10:41,000 So here you can see the red depicted as O's. 128 00:10:41,000 --> 00:10:50,240 So long story short, you've got these organic compounds and they come in various degrees 129 00:10:50,240 --> 00:10:57,900 of complexity and add a few more hydrogen molecules or carbon molecules and these can 130 00:10:57,900 --> 00:11:00,760 turn into two different substances. 131 00:11:00,760 --> 00:11:08,720 So I think that's also part of the magic that's happening behind the curtain in that people 132 00:11:08,720 --> 00:11:14,200 may develop slightly different pesticides. 133 00:11:14,200 --> 00:11:23,520 So they may develop a pesticide that's similar to trifloxistrobin, but it's slightly different. 134 00:11:23,520 --> 00:11:26,680 They'll give it a different chemical name. 135 00:11:26,680 --> 00:11:32,960 Say trifloxistrobin's regulated, the new molecule is not. 136 00:11:32,960 --> 00:11:39,720 It can be a way to still use pesticides while avoiding detection. 137 00:11:39,720 --> 00:11:49,680 So I've heard pesticide screening be described as a cat and mouse gang where they, and it's 138 00:11:49,680 --> 00:11:52,240 not just cannabis cultivators. 139 00:11:52,240 --> 00:11:56,600 This is, I think, could be an issue in agriculture in general. 140 00:11:56,600 --> 00:12:01,700 People engaged in agriculture, we've talked about this in the past, contaminants are of 141 00:12:01,700 --> 00:12:03,120 top concern. 142 00:12:03,120 --> 00:12:12,400 So they're always chasing new innovative ways to fight off pests and disease and molds and 143 00:12:12,400 --> 00:12:13,400 fungi. 144 00:12:13,400 --> 00:12:16,760 And so they're coming up with new chemicals. 145 00:12:16,760 --> 00:12:25,600 These chemicals are then being studied by toxicologists and then regulators will put 146 00:12:25,600 --> 00:12:33,120 limits on these and then say they'll be screened for in cannabis and then people will start 147 00:12:33,120 --> 00:12:37,240 using a new substance if they're failing a lot. 148 00:12:37,240 --> 00:12:40,280 So this is just something to watch for. 149 00:12:40,280 --> 00:12:46,960 Okay, so we haven't done too much yet except just describe the data. 150 00:12:46,960 --> 00:12:49,440 And that's often a useful first step. 151 00:12:49,440 --> 00:12:55,380 For example, this data, it's all publicly available and sitting right there in front 152 00:12:55,380 --> 00:12:56,380 of us. 153 00:12:56,380 --> 00:13:05,440 However, it took a good bit of curation simply to be able to count the number of pesticides 154 00:13:05,440 --> 00:13:08,240 that were detected. 155 00:13:08,240 --> 00:13:17,160 Well, not only can we count them, but we also know what type of product. 156 00:13:17,160 --> 00:13:25,720 So right now this isn't differentiating between say flour and hydrocarbon concentrates. 157 00:13:25,720 --> 00:13:34,680 Well, I'm going to show you a nifty way that you can model this data. 158 00:13:34,680 --> 00:13:44,600 Okay, so we've talked about how cultivators may want to view the risk, the chance of failing 159 00:13:44,600 --> 00:13:48,600 as a business risk and to factor that in. 160 00:13:48,600 --> 00:14:02,240 So for example, if you're cultivating flour and you value your flour at, I forget what 161 00:14:02,240 --> 00:14:08,560 the going rate is, but it's $1,000 a pound or $2,000 a pound. 162 00:14:08,560 --> 00:14:13,600 So that's your value of a pound of flour. 163 00:14:13,600 --> 00:14:17,040 I think they changed the flour lot sizes. 164 00:14:17,040 --> 00:14:23,360 And in fact, I saved them, but the old ones were a five pound lot. 165 00:14:23,360 --> 00:14:25,000 And so I think they changed them. 166 00:14:25,000 --> 00:14:27,920 I think some lots are up to 50 pounds now. 167 00:14:27,920 --> 00:14:33,080 But let's say you've got a five pound lot. 168 00:14:33,080 --> 00:14:41,320 Each pound will be parsimonious and say each pound is valued to you at $1,000. 169 00:14:41,320 --> 00:14:49,640 So a five pound lot has a value to your cultivator of $5,000. 170 00:14:49,640 --> 00:14:51,600 Okay. 171 00:14:51,600 --> 00:15:01,280 Well, if we were just going to do a simple count here, we know, let's just do a back 172 00:15:01,280 --> 00:15:03,880 of the envelope here real quick, right? 173 00:15:03,880 --> 00:15:07,200 So there's 600 give or take. 174 00:15:07,200 --> 00:15:20,000 Actually, let's find the exact number just to pesticides and perennial butoxide. 175 00:15:20,000 --> 00:15:21,000 Okay. 176 00:15:21,000 --> 00:15:22,000 644. 177 00:15:22,000 --> 00:15:23,000 Okay. 178 00:15:23,000 --> 00:15:32,960 Well, there's only, well, not only there's 48,000 observations and 644 of them failed 179 00:15:32,960 --> 00:15:35,400 for piperennial butoxide. 180 00:15:35,400 --> 00:15:41,760 So if you were just going to do a simple back of the envelope calculation, you could just 181 00:15:41,760 --> 00:15:46,240 say, oh, well, I know how many failed for piperennial butoxide. 182 00:15:46,240 --> 00:15:48,560 I know the total amount tested. 183 00:15:48,560 --> 00:15:57,220 Well, that's around 1.3% that are failing for piperennial butoxide. 184 00:15:57,220 --> 00:16:08,360 So if you have a $5,000 value flower lot and your risk is $66. 185 00:16:08,360 --> 00:16:16,960 So that's a $66 cost, expected cost of pesticide screening. 186 00:16:16,960 --> 00:16:26,120 And so $66 out of 5,000, right? 187 00:16:26,120 --> 00:16:29,020 We backed out the failure rate there. 188 00:16:29,020 --> 00:16:36,760 So pesticide screening imposes around a 1.3% cost. 189 00:16:36,760 --> 00:16:38,240 This is an unconditional cost. 190 00:16:38,240 --> 00:16:40,400 We haven't conditioned it on anything. 191 00:16:40,400 --> 00:16:43,620 We haven't conditioned on the product type. 192 00:16:43,620 --> 00:16:48,360 We have conditioned on where the grower may be located. 193 00:16:48,360 --> 00:16:51,460 So this is completely unconditional. 194 00:16:51,460 --> 00:17:00,280 The cultivator has an unconditional risk of 1.3% if they're growing a $5,000 lot. 195 00:17:00,280 --> 00:17:03,200 That's a $66 cost. 196 00:17:03,200 --> 00:17:10,240 It may not seem large, but margins are tight in the cannabis industry. 197 00:17:10,240 --> 00:17:12,320 And this adds up. 198 00:17:12,320 --> 00:17:21,200 So for example, if all of a sudden maybe you're selling not just $5,000 worth of product, 199 00:17:21,200 --> 00:17:30,960 but maybe you're selling $500,000 worth of product here, well, all of a sudden, now that's 200 00:17:30,960 --> 00:17:34,840 a $6,500 cost. 201 00:17:34,840 --> 00:17:36,880 $6,500. 202 00:17:36,880 --> 00:17:39,200 That's non-negligible. 203 00:17:39,200 --> 00:17:51,320 So if, say, your revenue is $500,000, it's a decent sized cost. 204 00:17:51,320 --> 00:17:56,720 And if all of a sudden you're faced with this cost and it's going to be lumpy, so that's 205 00:17:56,720 --> 00:18:00,240 assuming this is all evened out. 206 00:18:00,240 --> 00:18:07,520 Well, this is going to be a lumpy cost in that you may not fail pesticide screening 207 00:18:07,520 --> 00:18:17,160 for a long time, and then one month you fail and you have to now bear that entire cost 208 00:18:17,160 --> 00:18:19,160 in that one month. 209 00:18:19,160 --> 00:18:26,120 So this is what I'm trying to drive home in that just want to bear in mind that there 210 00:18:26,120 --> 00:18:27,120 is a cost. 211 00:18:27,120 --> 00:18:30,960 And you can minimize this cost. 212 00:18:30,960 --> 00:18:33,880 So for example, this is unconditional. 213 00:18:33,880 --> 00:18:41,600 Maybe we can find some conditions that may lower your probability of failing pesticide 214 00:18:41,600 --> 00:18:42,600 testing. 215 00:18:42,600 --> 00:18:46,000 And maybe you can manage those conditions. 216 00:18:46,000 --> 00:18:55,440 In economics, we learn that profit maximization is equivalent to cost minimization. 217 00:18:55,440 --> 00:18:59,820 And that's what I often try to stress to people in business. 218 00:18:59,820 --> 00:19:08,240 So many times people are worried about maximizing their profit, just trying to get as most money, 219 00:19:08,240 --> 00:19:10,720 as most revenue as possible. 220 00:19:10,720 --> 00:19:14,040 And they don't take into consideration their costs. 221 00:19:14,040 --> 00:19:24,360 And I can see people run awry with this, run astray with this. 222 00:19:24,360 --> 00:19:32,800 And an effective way to achieve the same result would just be to have a laser focus on keeping 223 00:19:32,800 --> 00:19:38,480 your costs to a minimum while engaging in as much business as you can. 224 00:19:38,480 --> 00:19:39,480 Cool. 225 00:19:39,480 --> 00:19:41,600 So I have droned on enough about that. 226 00:19:41,600 --> 00:19:45,520 Let's get back to some charts and actually some cool statistics here. 227 00:19:45,520 --> 00:19:50,640 So now I'm going to introduce conditional statistics. 228 00:19:50,640 --> 00:19:54,440 And these are the two things I tell you with, right? 229 00:19:54,440 --> 00:19:58,360 Start with simple statistics. 230 00:19:58,360 --> 00:20:00,400 So these are counts. 231 00:20:00,400 --> 00:20:03,960 These are actually conditional counts. 232 00:20:03,960 --> 00:20:07,880 These are counts by analyte. 233 00:20:07,880 --> 00:20:10,280 So those are conditional statistics. 234 00:20:10,280 --> 00:20:13,840 A count, a total is a statistic. 235 00:20:13,840 --> 00:20:19,440 A count by an analyte is a conditional statistic. 236 00:20:19,440 --> 00:20:25,000 They're amazingly powerful, interesting, and simple. 237 00:20:25,000 --> 00:20:28,440 And then you can just keep adding conditions. 238 00:20:28,440 --> 00:20:35,680 Well, as I alluded to earlier, perhaps product type matters. 239 00:20:35,680 --> 00:20:39,080 Well, how are we going to model this? 240 00:20:39,080 --> 00:20:49,360 The way I've encouraged people to go about modeling simple data in the past is binary. 241 00:20:49,360 --> 00:20:51,120 Zero, one. 242 00:20:51,120 --> 00:20:56,960 This is really going back to the basics in statistics. 243 00:20:56,960 --> 00:21:01,720 I'll be loading all the binary statistics episodes shortly. 244 00:21:01,720 --> 00:21:03,120 So those will be coming. 245 00:21:03,120 --> 00:21:10,880 This is just a clever way of modeling the data that opens it up for clever statistical 246 00:21:10,880 --> 00:21:11,880 exercises. 247 00:21:11,880 --> 00:21:12,880 OK. 248 00:21:12,880 --> 00:21:14,480 So what are we looking at here? 249 00:21:14,480 --> 00:21:24,840 So we're seeing that almost a little less than 48,000 of these samples had no piperanial 250 00:21:24,840 --> 00:21:27,920 butoxide detected. 251 00:21:27,920 --> 00:21:32,120 And a little more than something went awry in my counts here. 252 00:21:32,120 --> 00:21:34,720 But I'm just going to power on. 253 00:21:34,720 --> 00:21:40,760 And I want you to encourage you to come back through and double check my work like a good 254 00:21:40,760 --> 00:21:41,760 scientist would. 255 00:21:41,760 --> 00:21:51,040 And so 600 of these are coded as one, meaning that piperanial butoxide was detected. 256 00:21:51,040 --> 00:21:52,840 Fantastic. 257 00:21:52,840 --> 00:22:03,480 Well, when you have simple binary data, zero or one, you can estimate a probit model. 258 00:22:03,480 --> 00:22:12,680 And this would basically predict the probability of being zero versus the probability of being 259 00:22:12,680 --> 00:22:20,080 one conditional on your explanatory variables. 260 00:22:20,080 --> 00:22:22,240 This is exactly what we want. 261 00:22:22,240 --> 00:22:32,720 We want to know the probability that piperanial butoxide will be detected conditional on the 262 00:22:32,720 --> 00:22:34,600 sample type. 263 00:22:34,600 --> 00:22:43,360 We already know that the unconditional probability of piperanial butoxide being detected is around 264 00:22:43,360 --> 00:22:44,360 1.3%. 265 00:22:44,360 --> 00:22:50,280 Well, what's the probability in flour? 266 00:22:50,280 --> 00:22:54,400 What's the probability in hydrocarbon concentrates? 267 00:22:54,400 --> 00:23:00,080 You could do this again by just taking another conditional count. 268 00:23:00,080 --> 00:23:04,800 I'm simply introducing the probit model to you as a new... 269 00:23:04,800 --> 00:23:10,920 Well, we've already discovered this tool, but it's another tool that you can use to 270 00:23:10,920 --> 00:23:14,040 calculate conditional probabilities. 271 00:23:14,040 --> 00:23:20,120 And it's a tool that we can load up with explanatory variables. 272 00:23:20,120 --> 00:23:26,760 And I will save the big takeaway for the end of the day. 273 00:23:26,760 --> 00:23:31,600 You'll tease that the magic is going to happen with those explanatory variables. 274 00:23:31,600 --> 00:23:36,200 Now, I'm isolating a subsample here. 275 00:23:36,200 --> 00:23:43,720 So essentially, just want to look at flour lots and hydrocarbon concentrates. 276 00:23:43,720 --> 00:23:44,920 Why? 277 00:23:44,920 --> 00:23:46,880 Just to keep things simple. 278 00:23:46,880 --> 00:23:51,960 There's no reason why we couldn't look at more inventory types. 279 00:23:51,960 --> 00:23:57,440 These were just the two most frequent inventory types. 280 00:23:57,440 --> 00:24:00,480 And so just wanted to use them for a demonstration. 281 00:24:00,480 --> 00:24:02,840 Okay, enough of me rambling on. 282 00:24:02,840 --> 00:24:04,720 Let's estimate this model. 283 00:24:04,720 --> 00:24:07,800 So it's real simple to do. 284 00:24:07,800 --> 00:24:13,960 Well, and this is actually why I encourage you, if you really want to cut your teeth 285 00:24:13,960 --> 00:24:23,040 programming and learn statistics as intimately as possible, then I would challenge you to 286 00:24:23,040 --> 00:24:26,080 code a probit model. 287 00:24:26,080 --> 00:24:31,760 Because here we're just using one from the stats model package. 288 00:24:31,760 --> 00:24:34,640 And we can do it in one line of code. 289 00:24:34,640 --> 00:24:41,040 That's phenomenal, because now we can add this to our favorite programming algorithms 290 00:24:41,040 --> 00:24:48,320 and we could calculate thousands upon thousands of probit models in a short fashion. 291 00:24:48,320 --> 00:24:52,080 But this is a useful exercise. 292 00:24:52,080 --> 00:24:58,360 You'll have to hit, say Wikipedia or your favorite statistics textbook. 293 00:24:58,360 --> 00:25:08,560 And it's a non-trivial task, but it's completely in the realm of possible. 294 00:25:08,560 --> 00:25:13,600 And if you're up for the challenge, it's more of an exercise, right? 295 00:25:13,600 --> 00:25:21,640 Because at the end of the day, I would recommend to use the stats models, probit model, because 296 00:25:21,640 --> 00:25:27,920 one thing I love about open source is when it's used by many, many thousands and thousands, 297 00:25:27,920 --> 00:25:33,600 perhaps millions of people, then it becomes robust. 298 00:25:33,600 --> 00:25:40,160 And perhaps little quirks get discovered that you couldn't figure out by yourself. 299 00:25:40,160 --> 00:25:46,160 So probably the stats model's probit model would probably be more robust than any probit 300 00:25:46,160 --> 00:25:48,480 model that you may be able to write. 301 00:25:48,480 --> 00:25:51,120 You could probably write a better one, I'm sure. 302 00:25:51,120 --> 00:25:55,120 But as I said, it's a good exercise in both programming and statistics. 303 00:25:55,120 --> 00:25:56,840 And you can then double check it. 304 00:25:56,840 --> 00:26:01,040 You can basically double check that the probit model you wrote is equivalent. 305 00:26:01,040 --> 00:26:03,720 So that's just a tangent on programming. 306 00:26:03,720 --> 00:26:04,840 Here it is. 307 00:26:04,840 --> 00:26:10,520 And I'll just talk about the model and then show you the actual takeaway. 308 00:26:10,520 --> 00:26:18,720 So when you estimate a probit model, OK, so we've got our 30,000 observations. 309 00:26:18,720 --> 00:26:30,320 Our dependent variable, remember, that is going to be a binary, 0 or 1, mostly zeros. 310 00:26:30,320 --> 00:26:32,600 And in fact, that's actually a problem. 311 00:26:32,600 --> 00:26:35,920 And there may be better models out there. 312 00:26:35,920 --> 00:26:46,160 That's often a criticism in that if your zeros and ones aren't balanced, then that can lead 313 00:26:46,160 --> 00:26:47,160 to issues. 314 00:26:47,160 --> 00:26:51,160 But this can at least give us a good start. 315 00:26:51,160 --> 00:26:53,240 OK, so what do we have here? 316 00:26:53,240 --> 00:26:56,760 Well, we've got various coefficients here. 317 00:26:56,760 --> 00:27:04,000 And don't read too much into the coefficients alone, because the probit model, you actually 318 00:27:04,000 --> 00:27:06,240 have to evaluate it. 319 00:27:06,240 --> 00:27:11,200 The one thing that you can see is that this has a positive sign on it. 320 00:27:11,200 --> 00:27:17,980 So right out of the gate, we're thinking, OK, hydrocarbons may actually be detecting. 321 00:27:17,980 --> 00:27:23,000 We may be detecting piperannual butoxide more often in hydrocarbon concentrates. 322 00:27:23,000 --> 00:27:25,920 OK, well, how much? 323 00:27:25,920 --> 00:27:27,480 Well, guess what? 324 00:27:27,480 --> 00:27:29,320 It's actually super easy. 325 00:27:29,320 --> 00:27:38,700 So the way you go about this is you find what's called marginal probability of detecting for 326 00:27:38,700 --> 00:27:42,360 one category, and then you would do it for the other. 327 00:27:42,360 --> 00:27:47,680 And if these weren't categorical variables, then you would use the mean. 328 00:27:47,680 --> 00:27:55,840 So you would say, oh, what would be the probability of, say, your regressor was number of plants. 329 00:27:55,840 --> 00:28:03,960 What would be the probability of failing for piperannual butoxide given the, say, average 330 00:28:03,960 --> 00:28:05,840 number of plants? 331 00:28:05,840 --> 00:28:08,880 And so you can find different marginal probabilities. 332 00:28:08,880 --> 00:28:11,760 OK, so enough of me talking. 333 00:28:11,760 --> 00:28:13,240 Let's look at the numbers here. 334 00:28:13,240 --> 00:28:16,720 And we've got, OK, what's the marginal probability of flour? 335 00:28:16,720 --> 00:28:23,840 Well, here, I'll just plot all of this out, because this way, we'll have a nice visualization 336 00:28:23,840 --> 00:28:25,200 while I'm talking about it. 337 00:28:25,200 --> 00:28:34,120 So all I did here was evaluate the model at the mean of each of the groups. 338 00:28:34,120 --> 00:28:39,880 So here, flour, we've just got a constant. 339 00:28:39,880 --> 00:28:43,560 And hydrocarbon concentrates is always 0. 340 00:28:43,560 --> 00:28:45,280 That's a dummy variable. 341 00:28:45,280 --> 00:28:50,160 And then, of course, hydrocarbon concentrates has a constant. 342 00:28:50,160 --> 00:28:52,840 And hydrocarbon concentrates is always 1. 343 00:28:52,840 --> 00:29:00,960 So really simple to calculate the marginal probabilities for dummy variables. 344 00:29:00,960 --> 00:29:07,360 As I said, it's a bit more complicated if you're, say, you want to find the average 345 00:29:07,360 --> 00:29:08,840 marginal probability. 346 00:29:08,840 --> 00:29:12,040 Or I may be butchering my language here. 347 00:29:12,040 --> 00:29:18,600 But if you want to find the marginal probability of a continuous variable, I want to say you've 348 00:29:18,600 --> 00:29:20,000 used the mean. 349 00:29:20,000 --> 00:29:26,080 So let's just go ahead and interpret the results here, because I think this is an interesting 350 00:29:26,080 --> 00:29:30,960 takeaway that's non-obvious. 351 00:29:30,960 --> 00:29:39,960 If you talk to a chemist, kind of wanted to bounce this off of our good friends over at 352 00:29:39,960 --> 00:29:46,000 MCR Labs to see if this may be in line with what they've observed. 353 00:29:46,000 --> 00:29:52,920 So if any of you want to reach out to MCR Labs and see, hey, is this what you've observed 354 00:29:52,920 --> 00:29:58,840 or something similar with piperanial butoxide in cannabis? 355 00:29:58,840 --> 00:30:07,480 Because I think this is an interesting insight that is non-obvious and requires a good bit 356 00:30:07,480 --> 00:30:15,280 of data curation as well as clever conditional statistics to get to. 357 00:30:15,280 --> 00:30:20,360 And I think these are the types of insights that people in the cannabis industry could 358 00:30:20,360 --> 00:30:27,440 really benefit from, and they may be completely oblivious to the fact that they can get their 359 00:30:27,440 --> 00:30:29,440 hands on these statistics. 360 00:30:29,440 --> 00:30:30,440 So what is this? 361 00:30:30,440 --> 00:30:40,740 So this is now the conditional probability of piperanial butoxide being detected in your 362 00:30:40,740 --> 00:30:41,740 cannabis flower. 363 00:30:41,740 --> 00:30:49,680 Ooh, I also want to walk back that cost I was talking about earlier. 364 00:30:49,680 --> 00:30:57,120 This is just the probability of piperanial butoxide being detected. 365 00:30:57,120 --> 00:31:03,840 You may not necessarily fail quality assurance just because it's detected. 366 00:31:03,840 --> 00:31:14,240 So depending on what the cultivator's internal quality control is and the state limits, they 367 00:31:14,240 --> 00:31:22,800 may not actually face the 1.3% risk of losing the batch outright. 368 00:31:22,800 --> 00:31:27,280 But it at least gets you thinking in this ballpark. 369 00:31:27,280 --> 00:31:35,680 So this is sort of my main insight for today that, as I said, this may be apparent to chemists 370 00:31:35,680 --> 00:31:39,680 in the laboratory but may not be obvious to other people. 371 00:31:39,680 --> 00:31:48,720 Piperanial butoxide is detected at a higher rate in hydrocarbon concentrates than in flower 372 00:31:48,720 --> 00:31:49,720 lots. 373 00:31:49,720 --> 00:31:52,640 Well, why is this? 374 00:31:52,640 --> 00:32:03,640 One thing that's happening is, well, in hydrocarbon concentrates, yes, you are concentrating THC 375 00:32:03,640 --> 00:32:05,840 in other cannabinoids. 376 00:32:05,840 --> 00:32:15,920 Well, if you look at, and this is why the chemistry is mighty, mightily important. 377 00:32:15,920 --> 00:32:26,400 Look at this, the boiling point of piperanial butoxide is 356 degrees Fahrenheit. 378 00:32:26,400 --> 00:32:34,320 What is the boiling point of tetrahydrocannabinoid acid? 379 00:32:34,320 --> 00:32:37,640 I want to say we can find the boiling point, 180 degrees C. 380 00:32:37,640 --> 00:32:39,640 Okay, yes. 381 00:32:39,640 --> 00:32:41,320 So check this out. 382 00:32:41,320 --> 00:32:50,520 So THC boils at 155 degrees Celsius. 383 00:32:50,520 --> 00:32:54,960 So that's the point where THC turns into vapor. 384 00:32:54,960 --> 00:33:05,360 Well, if you're trying to keep THC as a liquid, a hydrocarbon concentrate, then you'd want 385 00:33:05,360 --> 00:33:12,080 to keep the temperature less than that when you're going through extraction. 386 00:33:12,080 --> 00:33:23,200 Well, if the temperature is less than 160 degrees Celsius at all times, piperanial butoxide 387 00:33:23,200 --> 00:33:26,760 does not boil off. 388 00:33:26,760 --> 00:33:35,000 So that means in this process, the idea is other compounds are being boiled off. 389 00:33:35,000 --> 00:33:44,440 So compounds you don't want, let's say you raise the temperature to 150 degrees Celsius. 390 00:33:44,440 --> 00:33:50,160 Well, a lot of the plant matter, I think, is going to start boiling off. 391 00:33:50,160 --> 00:33:52,920 You're going to have other solvents boil off. 392 00:33:52,920 --> 00:33:54,240 Well, guess what? 393 00:33:54,240 --> 00:33:57,760 Your piperanial butoxide will remain. 394 00:33:57,760 --> 00:34:07,040 And so what ends up happening is not only are you concentrating THC into hydrocarbon 395 00:34:07,040 --> 00:34:17,400 concentrates, you're also concentrating any high boiling point pesticides that may be 396 00:34:17,400 --> 00:34:18,400 in the product. 397 00:34:18,400 --> 00:34:25,760 So this may be a reason that concentrates have a higher detection rate. 398 00:34:25,760 --> 00:34:34,840 Another reason is perhaps flour that is more contaminated is more likely to be processed 399 00:34:34,840 --> 00:34:42,240 into concentrates than flour that's not. 400 00:34:42,240 --> 00:34:44,120 There's a lot going on here. 401 00:34:44,120 --> 00:34:52,520 However, I just wanted to point this out that if you are, say, a processor, then you may 402 00:34:52,520 --> 00:35:02,240 want to be a bit more concerned about piperanial butoxide detection than a cultivator. 403 00:35:02,240 --> 00:35:06,600 And just keep this in mind as you're factoring out your costs. 404 00:35:06,600 --> 00:35:13,800 As we pointed out, it's not the greatest cost in the world, but this is getting up there. 405 00:35:13,800 --> 00:35:21,320 This is now around, let's say you have to destroy your batch if it's detected, because 406 00:35:21,320 --> 00:35:23,440 maybe that's just your internal policy. 407 00:35:23,440 --> 00:35:27,600 Well, now you're facing that around 2% of the time. 408 00:35:27,600 --> 00:35:33,640 That's going to pose a non-negligible cost at that stage. 409 00:35:33,640 --> 00:35:36,680 And so just something to think about. 410 00:35:36,680 --> 00:35:42,000 And then basically, what's my big takeaway from this? 411 00:35:42,000 --> 00:35:49,360 This was really just an exercise, because as I said, you can reach out to MCR labs or 412 00:35:49,360 --> 00:35:55,960 your favorite chemists in the cannabis space, and chances are they'll be able to tell you, 413 00:35:55,960 --> 00:35:58,200 oh, yes, hydrocarbon concentrates. 414 00:35:58,200 --> 00:36:02,600 Yes, we detect pesticides at higher rates than in flour. 415 00:36:02,600 --> 00:36:04,760 Yes, piperanial butoxide. 416 00:36:04,760 --> 00:36:10,880 It may or may not be the most detected in other states, but chances are they may have 417 00:36:10,880 --> 00:36:11,880 seen that. 418 00:36:11,880 --> 00:36:19,080 So this may not be anything new under the sun, but this is a new tool that you can use. 419 00:36:19,080 --> 00:36:28,320 It's called the Pesticide Study for predicting the probability of pesticide failure. 420 00:36:28,320 --> 00:36:33,200 And that is exactly what I've been wanting to study for years. 421 00:36:33,200 --> 00:36:42,600 So probably since about 2018 or so, the same time this study came out, and it may not be 422 00:36:42,600 --> 00:36:51,400 2018, I've been definitely interested in what pesticides are being used in cannabis, 423 00:36:51,400 --> 00:36:59,120 what's the detection rate, what are the amounts being detected, and what are the factors that 424 00:36:59,120 --> 00:37:02,640 are causing pesticide contamination. 425 00:37:02,640 --> 00:37:10,120 And a hypothesis of mine that I've wanted to explore since 2018, but we just haven't 426 00:37:10,120 --> 00:37:22,520 had the data, is proximity to say agriculture or industry a potential factor in failing 427 00:37:22,520 --> 00:37:25,760 for pesticide testing? 428 00:37:25,760 --> 00:37:32,040 We have heard about pesticide drip in the past, and something that I'm interested in 429 00:37:32,040 --> 00:37:39,720 is let's say, is your risk of failing pesticides location dependent? 430 00:37:39,720 --> 00:37:46,000 So depending on where cultivators are in the state, or some of them at higher risk than 431 00:37:46,000 --> 00:37:50,720 others for pesticide screening. 432 00:37:50,720 --> 00:37:58,860 And so some of the conditions I wanted to look at were say, proximity to a dairy farm, 433 00:37:58,860 --> 00:38:01,120 or various types of farms. 434 00:38:01,120 --> 00:38:08,080 So people farm cherries and apples and all sorts of different commodities in Washington 435 00:38:08,080 --> 00:38:09,080 State. 436 00:38:09,080 --> 00:38:13,840 And what you could do is you could calculate those data points. 437 00:38:13,840 --> 00:38:19,600 So we know where the cultivators are with latitude and longitude, and then you could 438 00:38:19,600 --> 00:38:27,640 calculate the distance to say, the nearest dairy farm, or the nearest road, the nearest 439 00:38:27,640 --> 00:38:35,480 metropolitan area, and just try to look at all the different geographical factors that 440 00:38:35,480 --> 00:38:37,120 may be at play. 441 00:38:37,120 --> 00:38:41,020 And many of that may be insignificant. 442 00:38:41,020 --> 00:38:46,880 There may be no significant relationship on failure rate. 443 00:38:46,880 --> 00:38:51,680 But as I said, we're looking for something obvious here. 444 00:38:51,680 --> 00:38:55,560 And this piggybacks on the lesson of last week. 445 00:38:55,560 --> 00:39:02,920 When you're working with statistics, differences should hopefully be readily apparent pretty 446 00:39:02,920 --> 00:39:05,040 quickly out of the gate. 447 00:39:05,040 --> 00:39:11,920 And so here, pretty quickly out of the gate, we see, OK, hydrocarbon concentrates, flower 448 00:39:11,920 --> 00:39:12,920 lots. 449 00:39:12,920 --> 00:39:18,360 It looks like there's something structurally different about pesticide detection in these 450 00:39:18,360 --> 00:39:19,360 two. 451 00:39:19,360 --> 00:39:28,520 Well, now is the fun, because now we get to hunt for other explanatory variables that 452 00:39:28,520 --> 00:39:30,440 may be significant. 453 00:39:30,440 --> 00:39:37,120 And this is how you could potentially get a paper published, which would be quite an 454 00:39:37,120 --> 00:39:39,840 interesting feat. 455 00:39:39,840 --> 00:39:45,640 So I think last year, there was thousands of papers written on cannabis. 456 00:39:45,640 --> 00:39:53,400 And that shows you there's, one, a lot of interest and demand for this type of study. 457 00:39:53,400 --> 00:39:58,780 So perhaps if you think of a novel topic, then this is something that you could write 458 00:39:58,780 --> 00:40:01,600 about and that people could study. 459 00:40:01,600 --> 00:40:07,520 Or if you don't want to research, maybe you can think of a business model or a business 460 00:40:07,520 --> 00:40:11,440 endeavor to profit off of this information. 461 00:40:11,440 --> 00:40:20,320 So say you do figure out some structural risk for failing for pesticides. 462 00:40:20,320 --> 00:40:25,560 Well, maybe you could reach out to all the cultivators who are at high risk and just 463 00:40:25,560 --> 00:40:26,560 let them know. 464 00:40:26,560 --> 00:40:32,400 So you say, oh, maybe being close to a dairy farm puts you at high risk of failing for 465 00:40:32,400 --> 00:40:33,400 pesticide X. 466 00:40:33,400 --> 00:40:37,920 Well, you could reach out to all the cultivators near the dairy farm and just let them know, 467 00:40:37,920 --> 00:40:42,520 like, hey, you may want to be on the lookout for this pesticide. 468 00:40:42,520 --> 00:40:48,520 That's maybe not the best business endeavor ever, but I'll let you brainstorm and think 469 00:40:48,520 --> 00:40:49,520 of them. 470 00:40:49,520 --> 00:40:56,360 But long story short, that's what I have been wanting to study, as I said, for almost four 471 00:40:56,360 --> 00:40:58,000 years now. 472 00:40:58,000 --> 00:41:09,560 What are the geographic factors that may affect pesticide detection or failing for pesticides? 473 00:41:09,560 --> 00:41:10,960 There may be none, right? 474 00:41:10,960 --> 00:41:14,160 That's really a good null hypothesis, right? 475 00:41:14,160 --> 00:41:23,480 So the null hypothesis would be no geographic forces matter, right? 476 00:41:23,480 --> 00:41:28,520 Being close to a dairy farm doesn't matter, being close to a metropolitan area doesn't 477 00:41:28,520 --> 00:41:29,520 matter. 478 00:41:29,520 --> 00:41:35,940 So we would start with the null hypothesis of no effect and then see if we can't find 479 00:41:35,940 --> 00:41:42,440 evidence that any of these places may have a statistical effect on, say, failing for 480 00:41:42,440 --> 00:41:43,440 pesticides. 481 00:41:43,440 --> 00:41:48,000 So hopefully you found that interesting. 482 00:41:48,000 --> 00:41:54,080 I'm not sure if this is of widespread interest to people, but this was something that has 483 00:41:54,080 --> 00:41:58,920 been on my mind for a handful of years. 484 00:41:58,920 --> 00:42:04,220 And now we finally have the data that we can actually answer this question. 485 00:42:04,220 --> 00:42:06,360 So here's a start. 486 00:42:06,360 --> 00:42:11,000 We looked at one factor sample type. 487 00:42:11,000 --> 00:42:14,200 Well, now let's expand this, right? 488 00:42:14,200 --> 00:42:16,320 And that's the nice thing about the ProBit model. 489 00:42:16,320 --> 00:42:19,440 We can just pack in explanatory variables. 490 00:42:19,440 --> 00:42:21,920 So now is the fun time, right? 491 00:42:21,920 --> 00:42:30,440 And so this is, as I said, this is how statisticians have a fun conversation at lunch. 492 00:42:30,440 --> 00:42:38,600 You just start talking about, oh, you know, what explanatory variables may matter and 493 00:42:38,600 --> 00:42:40,120 do we have data on those? 494 00:42:40,120 --> 00:42:44,800 And if we don't have data on those, how can we proxy those or how can we measure that 495 00:42:44,800 --> 00:42:48,240 or how can we measure this? 496 00:42:48,240 --> 00:42:52,720 So now that you've got a good research question, you can start to brainstorm all the factors 497 00:42:52,720 --> 00:42:54,760 that are important. 498 00:42:54,760 --> 00:42:57,080 Go ahead and conclude it there. 499 00:42:57,080 --> 00:43:03,480 I feel like I'm rambling for the amount of material that we've covered. 500 00:43:03,480 --> 00:43:09,560 The goal, or not the goal, the insight of the day that I thought was fruitful. 501 00:43:09,560 --> 00:43:15,960 So last week's insight, I think, was helpful, but maybe not the most motivational ever. 502 00:43:15,960 --> 00:43:20,840 But as far as statistics goes, it was important. 503 00:43:20,840 --> 00:43:28,320 Well, this insight of the day is super motivational, and that is persist. 504 00:43:28,320 --> 00:43:29,320 Just keep trying. 505 00:43:29,320 --> 00:43:36,080 And it's going to be surprising how long some things take before they pay off. 506 00:43:36,080 --> 00:43:37,800 And what am I talking about? 507 00:43:37,800 --> 00:43:41,360 Well, I want to share with you some big news. 508 00:43:41,360 --> 00:43:45,000 So as you know, Candletics writes open source software. 509 00:43:45,000 --> 00:43:53,840 One of the tools that we've built was a open source software development kit for integrating 510 00:43:53,840 --> 00:43:57,360 with the metric API. 511 00:43:57,360 --> 00:44:04,840 And metric is the cannabis traceability software that is used in the vast majority of states 512 00:44:04,840 --> 00:44:07,520 that have legalized cannabis. 513 00:44:07,520 --> 00:44:18,280 And a long time ago, many moons ago, in 2021, so this was in March of 2021, I wrote this 514 00:44:18,280 --> 00:44:27,280 metric software development kit and got Candletics verified in a handful of states to interface 515 00:44:27,280 --> 00:44:28,840 with metric. 516 00:44:28,840 --> 00:44:35,280 Well, it takes a while for things to proliferate on the internet. 517 00:44:35,280 --> 00:44:40,760 And lately, the metric software development kit has been getting a lot of interest. 518 00:44:40,760 --> 00:44:48,640 And finally, Candletics has contracted to integrate a company with the metrics API. 519 00:44:48,640 --> 00:44:53,280 And this is really exciting, big news. 520 00:44:53,280 --> 00:44:57,240 One, just personally, it's a new contract. 521 00:44:57,240 --> 00:45:05,200 But two, this is a demonstration of this open source software being used in the wild. 522 00:45:05,200 --> 00:45:13,800 And I've teased a lot that, yes, I think this is software that you can use to make money. 523 00:45:13,800 --> 00:45:16,600 And I think I'm finally demonstrating that. 524 00:45:16,600 --> 00:45:23,800 So the contract is specifically just to help the person use this software. 525 00:45:23,800 --> 00:45:31,320 And this is one thing I learned from someone that if you show someone how to do something 526 00:45:31,320 --> 00:45:36,160 difficult, then often they'll pay you to do it. 527 00:45:36,160 --> 00:45:43,240 And so that's one of the reasons that we put out this software is we think this is a useful 528 00:45:43,240 --> 00:45:44,440 tool. 529 00:45:44,440 --> 00:45:46,720 We're happy to show people how to do it. 530 00:45:46,720 --> 00:45:51,600 And then if they need help using it, we're more than happy to help. 531 00:45:51,600 --> 00:45:57,960 So this was sort of big news I wanted to share with you and just let you know that, hey, 532 00:45:57,960 --> 00:46:06,000 this metric software development kit is actually quite useful on making money off of it. 533 00:46:06,000 --> 00:46:10,040 Hopefully this company gets value out of it. 534 00:46:10,040 --> 00:46:17,200 And as I was talking about with the Probit model earlier, the more and more people that 535 00:46:17,200 --> 00:46:20,400 use it, the more robust it becomes. 536 00:46:20,400 --> 00:46:27,440 So I would actually benefit if you use the open source software to make money, because 537 00:46:27,440 --> 00:46:33,480 hopefully if you use it and other companies use it, hopefully the software becomes more 538 00:46:33,480 --> 00:46:35,120 and more robust. 539 00:46:35,120 --> 00:46:43,080 Maybe you find something that needs to be ironed out and you can or cannot share if 540 00:46:43,080 --> 00:46:44,080 you wish. 541 00:46:44,080 --> 00:46:49,020 But if you decide to share, then hopefully I can improve the system. 542 00:46:49,020 --> 00:46:50,800 Hopefully that'll make your life better. 543 00:46:50,800 --> 00:46:52,520 Hopefully it'll make my life better. 544 00:46:52,520 --> 00:46:55,200 And we can just keep going back and forth. 545 00:46:55,200 --> 00:47:02,640 Just incremental progress, win-win, and value for value. 546 00:47:02,640 --> 00:47:06,280 It's taken a long time to get to this stage. 547 00:47:06,280 --> 00:47:10,600 And I want to thank you all for your support to get here. 548 00:47:10,600 --> 00:47:11,960 Every dollar mattered. 549 00:47:11,960 --> 00:47:17,920 I couldn't be here without literally each and every penny. 550 00:47:17,920 --> 00:47:22,560 Each and every penny got put to unbelievably good use. 551 00:47:22,560 --> 00:47:24,600 I want to thank you. 552 00:47:24,600 --> 00:47:31,120 Thank you for your eyes, your ears, your eyeballs contributing to CanLinux and the Canis Data 553 00:47:31,120 --> 00:47:33,000 Science Meetup Group. 554 00:47:33,000 --> 00:47:39,560 Because of you, we've covered an enormous amount of ground and I think are really starting 555 00:47:39,560 --> 00:47:46,600 to reap the benefits of the open source software and really start to provide people with value. 556 00:47:46,600 --> 00:47:53,760 So just want to give you that little bit of motivation that if you don't at first succeed, 557 00:47:53,760 --> 00:47:55,720 try, try, and try again. 558 00:47:55,720 --> 00:48:00,880 Also, I know it was me standing on the pedestal for most of the time today. 559 00:48:00,880 --> 00:48:04,000 So hopefully you at least found it interesting. 560 00:48:04,000 --> 00:48:10,480 Just in the last couple of minutes, do you have any thoughts, comments, questions, Candice, 561 00:48:10,480 --> 00:48:16,960 before we call it a start of the day? 562 00:48:16,960 --> 00:48:18,680 That is just fabulous. 563 00:48:18,680 --> 00:48:21,120 Fabulous presentation and coding. 564 00:48:21,120 --> 00:48:28,440 And that's interesting too, CanLinux is interfacing with the metric API. 565 00:48:28,440 --> 00:48:36,000 And Massachusetts uses this metric and also too, I do believe that MCR Labs, Yasha and 566 00:48:36,000 --> 00:48:45,880 Isaac will be very interested in really how you nail down that concentrates can also have 567 00:48:45,880 --> 00:48:47,920 concentrates of pesticides. 568 00:48:47,920 --> 00:48:48,920 That's concerning. 569 00:48:48,920 --> 00:48:56,320 As someone that did a lot of RSO in the beginning, I think that might fall under that product 570 00:48:56,320 --> 00:48:58,280 type category. 571 00:48:58,280 --> 00:49:05,640 And those are patients that are doing that RSO and we don't want high pesticides. 572 00:49:05,640 --> 00:49:12,380 So amazing Keegan, you're always uncovering just good stuff. 573 00:49:12,380 --> 00:49:14,200 So thank you for everything. 574 00:49:14,200 --> 00:49:20,560 I love your enthusiasm, Candice, and just to continue piggybacking on that, yes, CanLinux 575 00:49:20,560 --> 00:49:24,600 actually is verified with metric in Massachusetts. 576 00:49:24,600 --> 00:49:29,760 So that demonstrates that the software can be used in Massachusetts. 577 00:49:29,760 --> 00:49:37,640 And as you pointed out, this analysis hopefully will be quite valuable to people because personally, 578 00:49:37,640 --> 00:49:45,960 I just try to limit my consumption of foreign chemicals that I don't understand. 579 00:49:45,960 --> 00:49:54,040 So for me, everything else held constant, I would actually prefer the flower with no 580 00:49:54,040 --> 00:50:02,720 pesticides detected versus one with hyperannual butoxide detected, even if it is below the 581 00:50:02,720 --> 00:50:03,960 limit for the state. 582 00:50:03,960 --> 00:50:09,320 And so this is why I argue that if people work to make their lab results accessible 583 00:50:09,320 --> 00:50:14,080 to people, then this could actually be a selling point for you. 584 00:50:14,080 --> 00:50:20,800 People are always saying, what can we compete on besides THC concentration? 585 00:50:20,800 --> 00:50:23,560 Well, guess what? 586 00:50:23,560 --> 00:50:31,320 If there were two products, one was 18% THC, the other was 20% THC, and one clearly had 587 00:50:31,320 --> 00:50:39,040 on the label no pesticides detected, and the other one, hyperannual butoxide, was detected, 588 00:50:39,040 --> 00:50:43,360 it would be a no-brainer for me which one I would choose. 589 00:50:43,360 --> 00:50:45,560 This is a personal preference. 590 00:50:45,560 --> 00:50:48,240 Everybody has different preferences. 591 00:50:48,240 --> 00:50:54,800 If you're looking for a factor to compete on, then as a cultivator, you may want to really 592 00:50:54,800 --> 00:51:01,160 stress the fact if you don't have any pesticides detected in the flower that, yes, there's 593 00:51:01,160 --> 00:51:08,840 no pesticides detected in this sample or in this flower, and other cultivators may not 594 00:51:08,840 --> 00:51:15,720 necessarily be able to make that claim because as we were showing today, there's a non-negligible, 595 00:51:15,720 --> 00:51:17,560 and that was just one pesticide. 596 00:51:17,560 --> 00:51:19,240 There's other pesticides too. 597 00:51:19,240 --> 00:51:26,600 So a non-negligible amount are getting detections even if they may be below the limit. 598 00:51:26,600 --> 00:51:32,880 So say you're a concentrate provider, this could be a really big selling point for you. 599 00:51:32,880 --> 00:51:38,400 Well, also too, maybe you want to look for pesticides with a lower boiling point than 600 00:51:38,400 --> 00:51:40,000 THC, right? 601 00:51:40,000 --> 00:51:44,760 And just to get it chemically out of the product, right? 602 00:51:44,760 --> 00:51:51,440 Even though people might still want full disclosure that maybe no pesticides are used, but never 603 00:51:51,440 --> 00:51:52,440 mind. 604 00:51:52,440 --> 00:51:57,920 Well, you're thinking like a chemist, and one would hope that the processors, and I've 605 00:51:57,920 --> 00:52:01,760 met some really, really, really smart processors. 606 00:52:01,760 --> 00:52:05,440 I imagine a lot of them are taking this into consideration. 607 00:52:05,440 --> 00:52:13,280 However, we've talked in the past about asymmetric information exists, and we can't pretend it 608 00:52:13,280 --> 00:52:14,280 doesn't. 609 00:52:14,280 --> 00:52:21,240 So there may just be processors who aren't aware that boiling points of different pesticides 610 00:52:21,240 --> 00:52:22,560 matter. 611 00:52:22,560 --> 00:52:26,640 So one thing is just helping people access the information. 612 00:52:26,640 --> 00:52:31,000 So just do they even have the information to start with? 613 00:52:31,000 --> 00:52:34,040 That's a good piece. 614 00:52:34,040 --> 00:52:36,200 And then actionable insights. 615 00:52:36,200 --> 00:52:39,100 So we're working on those. 616 00:52:39,100 --> 00:52:44,120 We don't have too much that you can act on yet except taking into consideration the risk 617 00:52:44,120 --> 00:52:46,160 of flour versus concentrate. 618 00:52:46,160 --> 00:52:52,200 But I think there's a lot of doors that have been opened simply by creating this data. 619 00:52:52,200 --> 00:52:58,240 So now you can take this data, put it in a database, and do a quick query. 620 00:52:58,240 --> 00:53:05,120 So now all of a sudden, if you see a product, you can now know if there were any pesticides 621 00:53:05,120 --> 00:53:07,780 or say residual solvents detected. 622 00:53:07,780 --> 00:53:13,440 So I think this could be something that could help out consumers, could help out regulators, 623 00:53:13,440 --> 00:53:19,520 help out the labs, help out the cultivators and processors themselves, could even potentially 624 00:53:19,520 --> 00:53:21,120 help out retailers. 625 00:53:21,120 --> 00:53:24,160 Maybe retailers don't even know to look for this. 626 00:53:24,160 --> 00:53:27,180 So that's what we're all about. 627 00:53:27,180 --> 00:53:34,240 And that's what we love to do is find useful cannabis analytics that can help as many people 628 00:53:34,240 --> 00:53:39,040 as possible in the cannabis industry or the cannabis space. 629 00:53:39,040 --> 00:53:45,680 And as I said, this was a small start, but we're all about one molecule by a time. 630 00:53:45,680 --> 00:53:52,120 So literally today, we looked at yet another molecule, piperanial butoxide. 631 00:53:52,120 --> 00:53:57,280 And I think we moved the bowl forward, even if it was just one molecule. 632 00:53:57,280 --> 00:53:59,120 So I'm happy with today. 633 00:53:59,120 --> 00:54:00,120 So thank you. 634 00:54:00,120 --> 00:54:01,120 Thank you. 635 00:54:01,120 --> 00:54:03,440 Thank you for coming. 636 00:54:03,440 --> 00:54:10,480 Keep advancing cannabis data science.