1 00:00:00,000 --> 00:00:11,560 All right. 2 00:00:11,560 --> 00:00:14,080 Well, it's good to see everybody today. 3 00:00:14,080 --> 00:00:18,720 So welcome to the Cannabis Data Science Meetup group. 4 00:00:18,720 --> 00:00:19,720 My name is Keegan. 5 00:00:19,720 --> 00:00:22,120 I started a company, Canlytics. 6 00:00:22,120 --> 00:00:28,440 And so I've got a background working in a laboratory and have an economics degree, spent 7 00:00:28,440 --> 00:00:30,480 a bit of time as a data scientist. 8 00:00:30,480 --> 00:00:36,480 So trying to apply that to the cannabis field and try to share what I know and just help 9 00:00:36,480 --> 00:00:40,720 everybody else do cannabis analytics simply and easily. 10 00:00:40,720 --> 00:00:45,780 So just real quick, I guess we can go ahead and go around and just introduce ourselves 11 00:00:45,780 --> 00:00:46,780 real quick. 12 00:00:46,780 --> 00:00:49,920 So if you don't mind just taking 30 seconds real quick. 13 00:00:49,920 --> 00:00:52,080 So I'll just stop starting in the top corner. 14 00:00:52,080 --> 00:00:55,320 Charles, would you mind introducing yourself real quick? 15 00:00:55,320 --> 00:00:57,600 Hi, I'm Charles. 16 00:00:57,600 --> 00:01:04,680 I have 27 years of software development experience and I'm transitioning now into like data science 17 00:01:04,680 --> 00:01:07,680 and machine learning. 18 00:01:07,680 --> 00:01:09,680 Paul. 19 00:01:09,680 --> 00:01:10,680 Sure. 20 00:01:10,680 --> 00:01:13,200 Yeah, my name is Paul. 21 00:01:13,200 --> 00:01:18,840 I'm a data scientist with Gerald Motors. 22 00:01:18,840 --> 00:01:24,920 Been doing analytics work prior for about the last 10 to 12 years. 23 00:01:24,920 --> 00:01:29,800 I'm currently enrolled in a master's program at the University of Wisconsin, a master's 24 00:01:29,800 --> 00:01:31,720 degree in data science. 25 00:01:31,720 --> 00:01:34,600 Currently working on my final project for that program. 26 00:01:34,600 --> 00:01:38,800 And Charles and Keegan have been helping me out with my project. 27 00:01:38,800 --> 00:01:41,800 So welcome everybody. 28 00:01:41,800 --> 00:01:43,600 Awesome. 29 00:01:43,600 --> 00:01:48,400 Megan, it's good to see you here today. 30 00:01:48,400 --> 00:01:50,400 Would you mind introducing yourself? 31 00:01:50,400 --> 00:01:51,680 Hello. 32 00:01:51,680 --> 00:01:53,320 Nice to meet everybody. 33 00:01:53,320 --> 00:01:54,320 I'm Meg. 34 00:01:54,320 --> 00:01:58,720 I've been in data analytics for almost two years. 35 00:01:58,720 --> 00:02:03,600 I was just laid off from Cerner, but I was a business data analyst there. 36 00:02:03,600 --> 00:02:04,600 Awesome. 37 00:02:04,600 --> 00:02:07,800 Well, welcome to the group. 38 00:02:07,800 --> 00:02:11,360 It's always awesome to have more data scientists aboard. 39 00:02:11,360 --> 00:02:16,920 Arunabh, could you introduce yourself real quick if you don't mind? 40 00:02:16,920 --> 00:02:17,920 Sure, sure. 41 00:02:17,920 --> 00:02:21,360 Firstly, glad to meet everyone for the first time. 42 00:02:21,360 --> 00:02:22,800 And yeah, my name is Arunabh. 43 00:02:22,800 --> 00:02:28,800 Basically I'm a data science instructor at Lighthouse Labs, also looking to transition 44 00:02:28,800 --> 00:02:32,240 in the industry for any interesting opportunities. 45 00:02:32,240 --> 00:02:33,720 But yeah, that's what I do. 46 00:02:33,720 --> 00:02:36,920 Basically I've been in the field for four or five years now. 47 00:02:36,920 --> 00:02:37,920 Awesome. 48 00:02:37,920 --> 00:02:41,160 Well, it's great to have you. 49 00:02:41,160 --> 00:02:42,160 Ari. 50 00:02:42,160 --> 00:02:45,400 Hi, I'm Ari. 51 00:02:45,400 --> 00:02:50,400 I've been doing some web development and digital marketing, but recently took a data science 52 00:02:50,400 --> 00:02:56,360 bootcamp with General Assembly and looking to make transition into data science. 53 00:02:56,360 --> 00:02:57,360 Excellent. 54 00:02:57,360 --> 00:03:01,120 The more data scientists, the merrier. 55 00:03:01,120 --> 00:03:05,760 So Gabriel, if you don't mind. 56 00:03:05,760 --> 00:03:06,760 Hi. 57 00:03:06,760 --> 00:03:07,760 Oops. 58 00:03:07,760 --> 00:03:10,320 Sorry, had my webcam covered. 59 00:03:10,320 --> 00:03:12,760 I'm a data analytics student. 60 00:03:12,760 --> 00:03:16,920 I'm doing a bootcamp through Thinkful, so I'm pretty new to all this, but I'm open to 61 00:03:16,920 --> 00:03:18,920 learning as much as I can. 62 00:03:18,920 --> 00:03:21,680 Well, you're in the right place. 63 00:03:21,680 --> 00:03:24,220 So we're all learning here, including myself. 64 00:03:24,220 --> 00:03:26,920 So I've got some to learn today myself. 65 00:03:26,920 --> 00:03:29,800 So we've got Heather. 66 00:03:29,800 --> 00:03:32,160 It's up to you, Heather, but if you want to introduce yourself, you're more than welcome 67 00:03:32,160 --> 00:03:33,160 to. 68 00:03:33,160 --> 00:03:34,160 Hey, everybody. 69 00:03:34,160 --> 00:03:35,160 I'm Heather. 70 00:03:35,160 --> 00:03:39,280 I've been attending these meetings for a couple of weeks now. 71 00:03:39,280 --> 00:03:46,880 I don't have data science, but, or excuse me, data, but I hope to get a job soon where 72 00:03:46,880 --> 00:03:50,560 I'm actually collecting cannabis data and being able to analyze it. 73 00:03:50,560 --> 00:03:53,200 That would be more than exciting. 74 00:03:53,200 --> 00:04:00,640 So this meetup keeps me sane and looking forward to meeting you all. 75 00:04:00,640 --> 00:04:06,880 And how's everybody's 710? 76 00:04:06,880 --> 00:04:07,880 Productive and busy. 77 00:04:07,880 --> 00:04:10,360 So it's awesome to have you, Heather. 78 00:04:10,360 --> 00:04:11,360 Yeah. 79 00:04:11,360 --> 00:04:12,360 Yeah. 80 00:04:12,360 --> 00:04:13,720 The sales on 710 are way better than the 420. 81 00:04:13,720 --> 00:04:14,720 Just letting you all know. 82 00:04:14,720 --> 00:04:15,880 Oh, that's right. 83 00:04:15,880 --> 00:04:16,880 It's oil. 84 00:04:16,880 --> 00:04:18,400 It's a spill pack. 85 00:04:18,400 --> 00:04:19,400 Yes. 86 00:04:19,400 --> 00:04:21,600 So all the RSOs are like missing from the shelf. 87 00:04:21,600 --> 00:04:25,400 People just like take the whole one milliliter and dump it onto like some vanilla wafers. 88 00:04:25,400 --> 00:04:27,960 So you might not see RSOs for a while. 89 00:04:27,960 --> 00:04:30,960 They have to rejuvenate. 90 00:04:30,960 --> 00:04:33,960 Jesus. 91 00:04:33,960 --> 00:04:37,560 Interesting opportunity for some analytics there. 92 00:04:37,560 --> 00:04:38,840 Yeah, I like to do. 93 00:04:38,840 --> 00:04:42,880 So I like tearing into it if I can. 94 00:04:42,880 --> 00:04:47,960 Well, we may have to try to find some daily data and see if we can find a spike there. 95 00:04:47,960 --> 00:04:49,960 So awesome. 96 00:04:49,960 --> 00:04:57,440 Erin, last but not least, would you mind introducing yourself real quick? 97 00:04:57,440 --> 00:04:58,440 Getting a little bit ecstatic. 98 00:04:58,440 --> 00:04:59,440 That's not me. 99 00:04:59,440 --> 00:05:19,920 No, no, no. 100 00:05:19,920 --> 00:05:20,920 Keegan you're muted. 101 00:05:20,920 --> 00:05:28,600 Oh, I was saying, sorry Erin, I couldn't hear you and had to mute you for a second. 102 00:05:28,600 --> 00:05:33,360 But I'll try to re-mute. 103 00:05:33,360 --> 00:05:38,120 Oh, here we go. 104 00:05:38,120 --> 00:05:40,640 Yeah, I'm sorry. 105 00:05:40,640 --> 00:05:41,640 Just getting static. 106 00:05:41,640 --> 00:05:47,040 So we'll have to get your introduction here in a second unless you want to drop your drop 107 00:05:47,040 --> 00:05:53,600 a quick introduction for yourself in the chat and we'll revisit that. 108 00:05:53,600 --> 00:05:57,400 So it's awesome to have everyone here. 109 00:05:57,400 --> 00:05:59,680 So just to go ahead and kick it off. 110 00:05:59,680 --> 00:06:07,300 So each week we sort of look at an interesting data analytics question with the best cannabis 111 00:06:07,300 --> 00:06:09,500 data that we can find. 112 00:06:09,500 --> 00:06:14,580 So we've been primarily looking at lab results. 113 00:06:14,580 --> 00:06:19,240 So we have a data set in Washington state that Charles has been looking at trying to 114 00:06:19,240 --> 00:06:25,880 predict if we can, trying to see if we can predict when a sample may fail quality assurance 115 00:06:25,880 --> 00:06:36,100 testing for say microbial contamination, high residual solvents or potentially foreign matter. 116 00:06:36,100 --> 00:06:45,880 So it's tricky because most samples pass quality assurance screening and only a small number 117 00:06:45,880 --> 00:06:46,880 fail. 118 00:06:46,880 --> 00:06:56,720 So with limited variation it can lead to, it can be hard to fit a good model. 119 00:06:56,720 --> 00:07:06,720 So changing gears slightly, we found an interesting hemp, which is still cannabis, hemp data set 120 00:07:06,720 --> 00:07:14,720 from, let's see, I'm forgetting here the authors. 121 00:07:14,720 --> 00:07:23,140 So it's Midwestern hemp database from the University of Illinois. 122 00:07:23,140 --> 00:07:27,640 So I'll go ahead and share my screen real quick and just go ahead and get everybody 123 00:07:27,640 --> 00:07:32,360 up to speed here. 124 00:07:32,360 --> 00:07:35,160 Just to introduce Aaron real quick. 125 00:07:35,160 --> 00:07:40,760 He's a front end developer, also interested in cannabis data analysis. 126 00:07:40,760 --> 00:07:44,920 And so you're in the right spot as well here. 127 00:07:44,920 --> 00:07:50,480 So I'll share my screen real quick. 128 00:07:50,480 --> 00:08:03,840 So near the end we'll look at, I'll save some time so that way Charles can go over 129 00:08:03,840 --> 00:08:08,560 the work he's done because Charles has done some great work here predicting failures in 130 00:08:08,560 --> 00:08:09,920 Washington state. 131 00:08:09,920 --> 00:08:15,320 So I'll try to leave about 10 minutes or so at the end for Charles to present. 132 00:08:15,320 --> 00:08:20,520 So at least, so we'll try not to get too bogged down in this. 133 00:08:20,520 --> 00:08:25,600 So long story short, we found the Midwestern hemp database. 134 00:08:25,600 --> 00:08:40,400 And so, Philippa Burde is putting together data on hemp strains grown in Illinois, Wisconsin, 135 00:08:40,400 --> 00:08:43,680 Michigan, and Indiana. 136 00:08:43,680 --> 00:08:45,960 I think you can participate in other states. 137 00:08:45,960 --> 00:08:52,000 However, these are the cultivators that are currently participating. 138 00:08:52,000 --> 00:09:00,600 And so we've collected this data last week and have begun analyzing it. 139 00:09:00,600 --> 00:09:06,360 So what are some of the interesting data points we have here? 140 00:09:06,360 --> 00:09:15,120 Well, we have essentially the cultivar, which is the strain of cannabis grown. 141 00:09:15,120 --> 00:09:17,480 The source, we need to pin this down. 142 00:09:17,480 --> 00:09:20,560 It's either the company or the seed company. 143 00:09:20,560 --> 00:09:24,640 I believe it's actually the cultivator. 144 00:09:24,640 --> 00:09:28,520 And then we're interested in a couple variables here. 145 00:09:28,520 --> 00:09:35,520 So we're of course interested in what state it was grown in because we had an interesting 146 00:09:35,520 --> 00:09:45,000 observation last week that the soil breakdown of various states could affect the quality 147 00:09:45,000 --> 00:09:48,220 of hemp grown. 148 00:09:48,220 --> 00:09:54,800 And then we want to also look at the harvest date, if possible. 149 00:09:54,800 --> 00:10:01,600 So this is information about the cultivars, the strains. 150 00:10:01,600 --> 00:10:11,320 And so we ideally want to combine this with data they've provided on lab testing. 151 00:10:11,320 --> 00:10:16,480 So they've sent in each strain to get tested. 152 00:10:16,480 --> 00:10:24,840 And so they received results for CBD, THC, and CBG. 153 00:10:24,840 --> 00:10:27,120 Now I'm going to show you a graph below. 154 00:10:27,120 --> 00:10:39,720 So keep in mind the federal permitted level of THC in hemp is 0.3%. 155 00:10:39,720 --> 00:10:49,680 So if you accidentally grow hemp that's higher than 0.3%, then you need to essentially destroy 156 00:10:49,680 --> 00:10:51,360 your harvest. 157 00:10:51,360 --> 00:10:59,120 So we were talking last week about how it's a tricky dance that cultivators have to dance 158 00:10:59,120 --> 00:11:06,840 where they're trying to maximize the CBD in their products. 159 00:11:06,840 --> 00:11:16,720 However, as we can essentially visually see, there appears to be a positive correlation 160 00:11:16,720 --> 00:11:20,440 between CBD and THC. 161 00:11:20,440 --> 00:11:30,600 They're all cannabinoids deriving from CBG, CBGA to be precise. 162 00:11:30,600 --> 00:11:39,480 So as people try to maximize CBD in their hemp harvests, they run the risk of failing. 163 00:11:39,480 --> 00:11:45,000 So it comes down to the cultivator's risk preferences. 164 00:11:45,000 --> 00:11:54,960 However, there's calculus to be done to centrally try to maximize your CBD while keeping your 165 00:11:54,960 --> 00:12:00,760 probability of failing at a tolerable rate. 166 00:12:00,760 --> 00:12:09,160 So we're going to begin analysis here and see if we can't lend a hand and see if we 167 00:12:09,160 --> 00:12:14,440 can make any sense out of the data provided. 168 00:12:14,440 --> 00:12:29,320 So I have gone ahead and made this analysis available on GitHub. 169 00:12:29,320 --> 00:12:36,920 And I'll go ahead and push the latest commit so that way you can get your hands on the 170 00:12:36,920 --> 00:12:38,080 data. 171 00:12:38,080 --> 00:12:48,760 So essentially, I have copied and pasted this data into just an Excel spreadsheet just so 172 00:12:48,760 --> 00:12:52,000 that we can start to actually read the data. 173 00:12:52,000 --> 00:12:58,680 And so if you have a better way than copying and pasting, then have at it. 174 00:12:58,680 --> 00:13:04,720 However, that just was what I resorted to for the time being. 175 00:13:04,720 --> 00:13:18,360 So we've got the data. 176 00:13:18,360 --> 00:13:23,120 Then the analysis we're doing today will be available for you. 177 00:13:23,120 --> 00:13:27,640 And just to show you where that's located. 178 00:13:27,640 --> 00:13:38,280 So boosting all of the cannabis data science code to a GitHub repository. 179 00:13:38,280 --> 00:13:55,760 That way you can follow along, get the code and follow. 180 00:13:55,760 --> 00:14:05,560 So I must have accidentally saved this in last week's folder. 181 00:14:05,560 --> 00:14:07,320 So I'll fix this after the presentation. 182 00:14:07,320 --> 00:14:16,360 So if you are following along, the script is in the 07.07 folder at the time being. 183 00:14:16,360 --> 00:14:19,960 So long story short, we've got the data. 184 00:14:19,960 --> 00:14:24,680 So the first thing we need to do is look at the data. 185 00:14:24,680 --> 00:14:34,120 So well, simply I'm using Python, but you can use your programming language of choice. 186 00:14:34,120 --> 00:14:45,160 And so just going to... 187 00:14:45,160 --> 00:14:49,760 First things first, we want to just look at the data here. 188 00:14:49,760 --> 00:14:58,480 So I always just look at the first five observations, just as sort of a sanity check to see, okay, 189 00:14:58,480 --> 00:15:00,440 this is the data we have. 190 00:15:00,440 --> 00:15:08,240 And then if you're getting fancy, it's always a good idea just to generate some summary 191 00:15:08,240 --> 00:15:12,880 statistics of the data at hand. 192 00:15:12,880 --> 00:15:20,400 So we're working with about 750 hemp samples. 193 00:15:20,400 --> 00:15:33,120 So these were samples that were sent in by various cultivators and the strains. 194 00:15:33,120 --> 00:15:40,160 So what we can do is we can start adding some variables here. 195 00:15:40,160 --> 00:15:48,680 So the first thing I did was, okay, we can create a variable fail, which is just zero 196 00:15:48,680 --> 00:15:50,120 or one. 197 00:15:50,120 --> 00:16:00,920 And so it will be one if the hemp is higher than 0.3% concentration of THC and then zero 198 00:16:00,920 --> 00:16:06,720 otherwise. 199 00:16:06,720 --> 00:16:16,240 And let's look at that data. 200 00:16:16,240 --> 00:16:27,280 So we can see that about 25% of the samples are failing for high THC. 201 00:16:27,280 --> 00:16:30,680 So we can find that just by taking the mean. 202 00:16:30,680 --> 00:16:37,000 And we were talking about in Washington state how about only we were finding a shockingly 203 00:16:37,000 --> 00:16:38,560 low number, maybe less than 1%. 204 00:16:38,560 --> 00:16:43,080 Correct me if I'm wrong, Charles, but it seemed that it was low. 205 00:16:43,080 --> 00:16:47,800 Percent of samples fall failing for quality assurance reason. 206 00:16:47,800 --> 00:16:55,720 When you have such low variance, you typically have to resort to special models. 207 00:16:55,720 --> 00:17:02,200 So there's a negative binomial model, but in general analysis can be tricky. 208 00:17:02,200 --> 00:17:08,560 So what's interesting about the hemp data is what I would consider a non-negligible 209 00:17:08,560 --> 00:17:14,480 portion of the sample is failing for high THC. 210 00:17:14,480 --> 00:17:17,920 And you have a high standard deviation. 211 00:17:17,920 --> 00:17:21,600 So you've got a lot of variation here. 212 00:17:21,600 --> 00:17:25,500 So it may prove for rich analytics. 213 00:17:25,500 --> 00:17:36,600 So we may be able to help these hemp producers out or labs that are testing hemp. 214 00:17:36,600 --> 00:17:45,720 So this is where we got to last week where I was essentially saying you can find quite 215 00:17:45,720 --> 00:17:52,820 interesting statistics by just taking conditional averages. 216 00:17:52,820 --> 00:17:56,640 So here I'm just taking an average. 217 00:17:56,640 --> 00:18:03,600 So here I'm just taking the mean for each state. 218 00:18:03,600 --> 00:18:09,840 So I'll go ahead and print that out just so that you can see. 219 00:18:09,840 --> 00:18:16,400 So if you do a conditional average, you can see the failure rate varies a little bit from 220 00:18:16,400 --> 00:18:18,120 state to state. 221 00:18:18,120 --> 00:18:27,440 So Illinois has the highest failure rate at above 30% failing for high THC. 222 00:18:27,440 --> 00:18:33,860 And then Wisconsin has the lowest failure rate for high hemp at just over 20%. 223 00:18:33,860 --> 00:18:36,880 So there's slight variation state by state. 224 00:18:36,880 --> 00:18:42,640 However, it'd be interesting to see if there's anything to that. 225 00:18:42,640 --> 00:18:49,360 So for now, I've just taken the conditional average, and there's about a 10% range. 226 00:18:49,360 --> 00:18:52,680 So it may be worth exploring further. 227 00:18:52,680 --> 00:18:56,880 For example, soil. 228 00:18:56,880 --> 00:19:07,700 Next, just wanted to think about whether some of the other data points that we have here. 229 00:19:07,700 --> 00:19:11,440 So if we just print out our columns. 230 00:19:11,440 --> 00:19:14,440 So we have the state. 231 00:19:14,440 --> 00:19:19,600 I think it would be interesting to look at the county to see if you could find county 232 00:19:19,600 --> 00:19:22,280 level soil data. 233 00:19:22,280 --> 00:19:25,480 So there's a lot of factors you could add. 234 00:19:25,480 --> 00:19:28,840 You could do precipitation. 235 00:19:28,840 --> 00:19:30,120 You could be creative. 236 00:19:30,120 --> 00:19:35,080 So I think there's room for analysis there. 237 00:19:35,080 --> 00:19:37,400 Geographic analysis can take a little bit of time. 238 00:19:37,400 --> 00:19:45,320 So just for expediency sake, I began looking at centrally when the hemp was sampled. 239 00:19:45,320 --> 00:19:51,280 So we have the date when the hemp was sampled. 240 00:19:51,280 --> 00:19:57,840 So just looking at the data here, just the way the data was collected. 241 00:19:57,840 --> 00:20:01,160 And this is what you'll learn when you're doing data analytics. 242 00:20:01,160 --> 00:20:02,920 And a lot of you have already learned. 243 00:20:02,920 --> 00:20:05,960 You spend a lot of your time just cleaning up the data. 244 00:20:05,960 --> 00:20:12,840 So from the way I copied and pasted this in, you'll see there's these unnecessary spaces. 245 00:20:12,840 --> 00:20:15,280 The year is in its own column. 246 00:20:15,280 --> 00:20:17,480 So not ideal. 247 00:20:17,480 --> 00:20:28,920 So centrally, these lines here are just turning the date, the sample date, and the sample 248 00:20:28,920 --> 00:20:33,640 year into a nice time stamp. 249 00:20:33,640 --> 00:20:39,080 That way we can work with it as a time series. 250 00:20:39,080 --> 00:20:40,720 But we're not doing time series analysis. 251 00:20:40,720 --> 00:20:46,440 But we can at least have it in daytime. 252 00:20:46,440 --> 00:20:56,200 So once again, make sure to look at the data. 253 00:20:56,200 --> 00:21:02,720 OK. 254 00:21:02,720 --> 00:21:20,080 So what we have here is essentially the earliest sample date was July 27. 255 00:21:20,080 --> 00:21:28,360 And then the latest was November 16. 256 00:21:28,360 --> 00:21:36,680 I should have here we are. 257 00:21:36,680 --> 00:21:44,240 I think if we take the mean of OK. 258 00:21:44,240 --> 00:21:46,680 So I wrote this little function up here. 259 00:21:46,680 --> 00:21:52,620 So basically, the way I'm calculating this is I'm calculating this is the number of days 260 00:21:52,620 --> 00:21:59,800 into the year that sampling occurred. 261 00:21:59,800 --> 00:22:09,560 So for example, the median. 262 00:22:09,560 --> 00:22:25,280 I think we wanted integer. 263 00:22:25,280 --> 00:22:32,400 So perhaps we want strings. 264 00:22:32,400 --> 00:22:37,920 The average harvest occurred on September 19. 265 00:22:37,920 --> 00:22:42,020 So that was when the average harvest occurred. 266 00:22:42,020 --> 00:22:53,320 And then we saw that the latest harvest was in November, the earliest in July. 267 00:22:53,320 --> 00:22:59,440 So there's some variation between when people are harvesting their hemp. 268 00:22:59,440 --> 00:23:06,920 And so essentially, I was wanting to see when people sample their hemp. 269 00:23:06,920 --> 00:23:12,320 So keep in mind, we essentially want to combine some data sets here. 270 00:23:12,320 --> 00:23:16,040 So it would be nice to actually use the harvest date. 271 00:23:16,040 --> 00:23:25,000 However, there's not a one to one relationship here between the agronomic performance and 272 00:23:25,000 --> 00:23:27,260 the cannabinoid data. 273 00:23:27,260 --> 00:23:30,500 So I haven't connected the dots yet. 274 00:23:30,500 --> 00:23:34,200 So I'm using the sample date. 275 00:23:34,200 --> 00:23:43,680 And it's not clear the relation between sample date and harvest date. 276 00:23:43,680 --> 00:23:48,720 So the harvest date seems to be in the future from the sample date. 277 00:23:48,720 --> 00:23:51,840 That's what I had a sneaking suspicion of. 278 00:23:51,840 --> 00:23:59,040 However, it doesn't. 279 00:23:59,040 --> 00:24:01,320 That may just be the situation. 280 00:24:01,320 --> 00:24:09,680 So they may be sampling before they harvest, which would make sense. 281 00:24:09,680 --> 00:24:16,160 So that way, they can just be continuously monitoring their cannabinoids and then harvest 282 00:24:16,160 --> 00:24:21,680 before it's too late. 283 00:24:21,680 --> 00:24:31,680 I think we need to do a deeper dive here into this data to actually understand a bit more 284 00:24:31,680 --> 00:24:34,040 about these data points. 285 00:24:34,040 --> 00:24:43,840 So I'll go ahead and hedge that now that we need to understand sampling date and harvest 286 00:24:43,840 --> 00:24:48,280 date a little better. 287 00:24:48,280 --> 00:24:57,000 However, continuing just to keep in mind, a lot of what I'm doing now is essentially 288 00:24:57,000 --> 00:25:06,400 a demonstration, a crude demonstration of how you can start going about data analytics. 289 00:25:06,400 --> 00:25:10,640 So in practice, you'll want to do things. 290 00:25:10,640 --> 00:25:15,220 You want to research these data points a little better. 291 00:25:15,220 --> 00:25:19,300 Study your models a little better. 292 00:25:19,300 --> 00:25:21,220 Do a bit more background research. 293 00:25:21,220 --> 00:25:24,680 So just keep that in mind. 294 00:25:24,680 --> 00:25:32,280 This is just a quick crude analysis just to give you an idea of what you can do. 295 00:25:32,280 --> 00:25:38,040 Hey, Keegan, a quick question for you on that actual source data. 296 00:25:38,040 --> 00:25:42,880 So I notice it's called hemp data. 297 00:25:42,880 --> 00:25:47,520 I assume it's just another, they're using that term generically as just a cannabis crop, 298 00:25:47,520 --> 00:25:49,400 but are this grown outside? 299 00:25:49,400 --> 00:25:53,800 Are these particular plants grown outside or crops grown outside? 300 00:25:53,800 --> 00:25:55,560 Hemp is typically grown outside. 301 00:25:55,560 --> 00:26:03,480 So typically people just seed acres and acres of hemp. 302 00:26:03,480 --> 00:26:06,160 There's no rules one way or the other. 303 00:26:06,160 --> 00:26:12,040 So you can grow hemp in, say, a greenhouse or potentially indoors. 304 00:26:12,040 --> 00:26:16,520 However, there's a couple of things going on here. 305 00:26:16,520 --> 00:26:22,560 From my understanding, there's essentially three reasons people grow hemp. 306 00:26:22,560 --> 00:26:29,600 I'm sure there's more, but the three main ones are for CBD, so for processing the flower 307 00:26:29,600 --> 00:26:30,940 into CBD. 308 00:26:30,940 --> 00:26:36,840 So there you're going to be growing flower stalks like you see in this picture, and then 309 00:26:36,840 --> 00:26:40,360 you'll want to be harvesting that for CBD. 310 00:26:40,360 --> 00:26:42,680 The second reason is for fiber. 311 00:26:42,680 --> 00:26:46,440 And so then you're going to be growing much more fibrous hemp. 312 00:26:46,440 --> 00:26:51,120 And so there you typically won't flower quite like that, and they'll look a lot more like 313 00:26:51,120 --> 00:26:52,880 stalks. 314 00:26:52,880 --> 00:26:58,360 And then the third reason is for seeds, for hemp seeds. 315 00:26:58,360 --> 00:27:05,080 And so then you'll typically have something that you'd have in this field, but then you'll 316 00:27:05,080 --> 00:27:06,080 have them pollinated. 317 00:27:06,080 --> 00:27:12,720 So you'll have a bunch of male plants in there with the females, and they'll produce seeds. 318 00:27:12,720 --> 00:27:19,600 And then people can either eat hemp seeds or hemp seeds are actually quite valuable 319 00:27:19,600 --> 00:27:20,680 these days. 320 00:27:20,680 --> 00:27:27,240 So that's something that I'm kind of coming to understand is there's such a high demand 321 00:27:27,240 --> 00:27:32,360 for cultivation that the cultivators need to get seeds. 322 00:27:32,360 --> 00:27:39,960 So there's actually it's actually quite profitable to grow seeds is what I am coming to understand. 323 00:27:39,960 --> 00:27:40,960 I gotcha. 324 00:27:40,960 --> 00:27:42,520 Yeah, thanks for explaining that. 325 00:27:42,520 --> 00:27:47,640 I wasn't quite up on the hemp side of the house, so that helps. 326 00:27:47,640 --> 00:27:48,640 Exactly. 327 00:27:48,640 --> 00:27:51,000 So it's a bit more of an agricultural crop. 328 00:27:51,000 --> 00:27:57,720 So it's not the exact same as high THC cannabis. 329 00:27:57,720 --> 00:28:05,680 So is that hence the point three percent or the point three percent threshold for THC? 330 00:28:05,680 --> 00:28:06,680 Point oh three maybe. 331 00:28:06,680 --> 00:28:07,680 It's a point oh three. 332 00:28:07,680 --> 00:28:08,680 It's like nothing. 333 00:28:08,680 --> 00:28:11,560 Yeah, it's point three percent. 334 00:28:11,560 --> 00:28:16,200 So it's it's a shockingly low percent of THC. 335 00:28:16,200 --> 00:28:23,140 However, when you really dive into the science, it is still just the cannabis plant. 336 00:28:23,140 --> 00:28:28,960 So it just some varieties just don't produce much THC. 337 00:28:28,960 --> 00:28:36,200 And so it's real interesting because you'll end. 338 00:28:36,200 --> 00:28:42,760 You know, I worked in a lab that tested hemp products and you'll find farmers who they'll 339 00:28:42,760 --> 00:28:49,880 basically start growing hemp and they think they may have a low THC variety and it may 340 00:28:49,880 --> 00:28:57,880 produce phenomenal CBD, but sometimes it may have like seven percent THC and then they 341 00:28:57,880 --> 00:29:01,640 will have to destroy their harvest. 342 00:29:01,640 --> 00:29:08,720 So it's tricky, right? 343 00:29:08,720 --> 00:29:15,520 Because of course, you know, there are just these strains that definitely don't produce 344 00:29:15,520 --> 00:29:23,400 much THC, but as we can kind of see here, there's. 345 00:29:23,400 --> 00:29:26,600 You know, it's not like they don't produce THC. 346 00:29:26,600 --> 00:29:33,760 So it's basically just people have selected the cannabis for different purposes. 347 00:29:33,760 --> 00:29:44,720 So people selling high THC flour or or oil, they've basically just they just, you know, 348 00:29:44,720 --> 00:29:51,480 selected the plants that have the high THC and they've just bred for high THC. 349 00:29:51,480 --> 00:29:59,800 And that's why you see strains in the stores with, you know, quite high levels of THC, 350 00:29:59,800 --> 00:30:02,640 like above 20 percent THC. 351 00:30:02,640 --> 00:30:06,920 So it's really just what people are selecting for. 352 00:30:06,920 --> 00:30:15,680 And so here people are trying to select like high CBD strains that still don't produce 353 00:30:15,680 --> 00:30:16,680 much THC. 354 00:30:16,680 --> 00:30:17,680 Gotcha. 355 00:30:17,680 --> 00:30:18,680 Makes sense. 356 00:30:18,680 --> 00:30:32,120 So just to kind of wrap this up, what we'll ultimately want to do is somehow analyze the 357 00:30:32,120 --> 00:30:39,440 cultivars, because that's what from my analysis is that's what I try to kind of boil this 358 00:30:39,440 --> 00:30:43,360 down to, because basically I'll get to it at the end. 359 00:30:43,360 --> 00:30:49,320 But I was thinking that, OK, these. 360 00:30:49,320 --> 00:30:55,200 The cultivator is choosing, OK, this is when I harvest, this is when I sample. 361 00:30:55,200 --> 00:31:02,160 But then it may it's in a way, in a way, it's almost the plant deciding, right, because 362 00:31:02,160 --> 00:31:07,880 really the cultivator is just essentially waiting for the plant to finish the flower, 363 00:31:07,880 --> 00:31:12,480 to seed and then they'll they'll they'll harvest it when it's ready. 364 00:31:12,480 --> 00:31:22,640 So the choice is more in choosing the strain, the cultivar, but that has the your ideal 365 00:31:22,640 --> 00:31:23,640 traits. 366 00:31:23,640 --> 00:31:32,160 So choosing a cultivar that that finishes harvest in September. 367 00:31:32,160 --> 00:31:38,120 So it's more the cultivar is finishing in September, whereas some cultivars finish in 368 00:31:38,120 --> 00:31:39,120 November. 369 00:31:39,120 --> 00:31:40,120 But. 370 00:31:40,120 --> 00:31:44,600 But that just threw thought through the time being. 371 00:31:44,600 --> 00:31:50,560 So let's just go ahead and get back to the analysis here. 372 00:31:50,560 --> 00:31:57,800 So just to to walk through just some some basic, some simple regressions just to continue 373 00:31:57,800 --> 00:32:00,480 looking at the data. 374 00:32:00,480 --> 00:32:11,120 Essentially just took a regression of total THC on the days into the year that that the 375 00:32:11,120 --> 00:32:13,720 hemp was sampled. 376 00:32:13,720 --> 00:32:19,720 And so as you would expect, there's a slight positive coefficient. 377 00:32:19,720 --> 00:32:26,480 So the later you wait, the higher the THC is going to be. 378 00:32:26,480 --> 00:32:37,640 And similarly, if you run a regression of total CBD on the days into the year where 379 00:32:37,640 --> 00:32:46,680 sampling occurred, you'll find another positive coefficient. 380 00:32:46,680 --> 00:32:53,760 So this was what we were beginning to talk about last week, where it appears that the 381 00:32:53,760 --> 00:33:00,040 longer into the year you wait to harvest or the longer into the year the plant takes to 382 00:33:00,040 --> 00:33:04,760 harvest, the higher the cannabinoids are going to be. 383 00:33:04,760 --> 00:33:13,640 So we were wondering, does that mean your chance of failing is going to be higher? 384 00:33:13,640 --> 00:33:20,960 So introducing you today, well, a lot of you may have already seen it, is the logistic 385 00:33:20,960 --> 00:33:22,440 regression. 386 00:33:22,440 --> 00:33:32,000 And so I put some some sources here at the top so that way you can just begin. 387 00:33:32,000 --> 00:33:40,400 There's everything from just some tutorials to some originals. 388 00:33:40,400 --> 00:33:50,840 So for example, so if you need some more in-depth reading, there it is for you. 389 00:33:50,840 --> 00:33:54,960 And then essentially, you just need a light tutorial. 390 00:33:54,960 --> 00:33:57,520 There's Geeks for Geeks. 391 00:33:57,520 --> 00:34:04,360 So I've got a couple references here for you to read up on logistic regressions. 392 00:34:04,360 --> 00:34:14,800 So long story short, I wish there was a better picture here. 393 00:34:14,800 --> 00:34:22,960 Well, basically, your regression can take a value of zero or one. 394 00:34:22,960 --> 00:34:32,900 And so they're useful for predicting binary outcomes because you'll basically run a regression 395 00:34:32,900 --> 00:34:39,520 if your estimate is less than 0.5, then your prediction is zero. 396 00:34:39,520 --> 00:34:45,800 And then if your prediction is greater than 0.5, your prediction is one. 397 00:34:45,800 --> 00:34:54,480 So anything you can dichotomize as zero and one, and you have some explanatory variables, 398 00:34:54,480 --> 00:34:57,600 then you can run a logistic regression. 399 00:34:57,600 --> 00:35:00,900 And there's a lot more to it. 400 00:35:00,900 --> 00:35:08,640 So it depends on whether your explanatory variables depends on the choice or on the 401 00:35:08,640 --> 00:35:09,640 individual. 402 00:35:09,640 --> 00:35:15,600 So I'll let you dive into the reading, but I won't bog you down with that right now. 403 00:35:15,600 --> 00:35:19,120 And we'll just get to the analysis. 404 00:35:19,120 --> 00:35:30,840 Long story short, remember, we have coded a failure variable, which is 405 00:35:30,840 --> 00:35:35,000 just zero if it passes. 406 00:35:35,000 --> 00:35:44,240 And so that would be everything below the red line is zero, everything above the red 407 00:35:44,240 --> 00:35:46,640 line is one. 408 00:35:46,640 --> 00:35:53,960 And so basically, we're going to see if we can predict if things, if the hemp sample 409 00:35:53,960 --> 00:35:54,960 fails. 410 00:35:54,960 --> 00:36:01,600 So we've got things coded, and we're just going to use the sample that date. 411 00:36:01,600 --> 00:36:09,120 So does the date, the time into the year when things were sampled, can that be used to predict 412 00:36:09,120 --> 00:36:14,160 when things were, when hemp fails? 413 00:36:14,160 --> 00:36:20,920 So you can basically fit a logistic regression. 414 00:36:20,920 --> 00:36:27,880 Regression doesn't mean too too much, except for if it's positive or negative. 415 00:36:27,880 --> 00:36:37,400 So long story short, let's go ahead and essentially look at the predictions. 416 00:36:37,400 --> 00:36:39,720 The predictions are the most interesting. 417 00:36:39,720 --> 00:36:45,360 Well, that may be my opinion, but it's sort of where the analysis lies with the logistic 418 00:36:45,360 --> 00:36:47,000 regression. 419 00:36:47,000 --> 00:36:58,200 So essentially, using the regression line that we fit, I predicted the probability of 420 00:36:58,200 --> 00:37:07,480 failing for high THC given the days into the year when sampling occurred. 421 00:37:07,480 --> 00:37:20,040 So keep in mind, so here's another plot, which basically adds the, what we observed. 422 00:37:20,040 --> 00:37:26,400 And so I'll be frank, it's not really the best fit. 423 00:37:26,400 --> 00:37:34,680 So when you see in the examples and tutorials out there, logistic regressions, you'll see 424 00:37:34,680 --> 00:37:37,600 much better fits than this. 425 00:37:37,600 --> 00:37:47,400 So graphically, it doesn't look like the best fit, but you can essentially start to see, 426 00:37:47,400 --> 00:37:55,920 keep in mind this is a crude analysis, but essentially what the prediction is, is the 427 00:37:55,920 --> 00:38:05,600 later into the year that you wait, the chance of failing for high THC increases. 428 00:38:05,600 --> 00:38:16,280 So all the way, so this is why I have this function right here. 429 00:38:16,280 --> 00:38:26,080 Oh, well, I guess I've got this plotted down here. 430 00:38:26,080 --> 00:38:35,880 It looks like all the way through September or so, you only have maybe less than a 25% 431 00:38:35,880 --> 00:38:38,000 chance of failing. 432 00:38:38,000 --> 00:38:48,200 And then it looks like, as you wait, your chances of failing increases. 433 00:38:48,200 --> 00:38:55,600 So I wanted to quantify this a little bit better. 434 00:38:55,600 --> 00:39:02,760 So using just another statistical package. 435 00:39:02,760 --> 00:39:09,840 So here I use stats models, and then here I used scikit-learn. 436 00:39:09,840 --> 00:39:15,880 So once again, it's just a package. 437 00:39:15,880 --> 00:39:22,040 In fact, I actually have an implementation right here that if you want to see how to 438 00:39:22,040 --> 00:39:32,680 write a logistic regression with just matrices, then have at it. 439 00:39:32,680 --> 00:39:38,520 So long story short, I'm just using tools here. 440 00:39:38,520 --> 00:39:48,880 Essentially trying to just quantify this regression to see, well, how accurate was it? 441 00:39:48,880 --> 00:40:02,360 So apparently it predicted with 76% accuracy, but I just kind of want to show the breakdown 442 00:40:02,360 --> 00:40:05,280 of that. 443 00:40:05,280 --> 00:40:13,720 And so here we're predicting failures, and it's actually a failure. 444 00:40:13,720 --> 00:40:17,360 So that happened 16 times. 445 00:40:17,360 --> 00:40:22,600 Then here we predicted a failure, and it's actually a pass. 446 00:40:22,600 --> 00:40:25,000 That's 12 times. 447 00:40:25,000 --> 00:40:31,960 Here we predicted a pass, and it actually failed 168 times. 448 00:40:31,960 --> 00:40:33,600 And I'm going to revisit. 449 00:40:33,600 --> 00:40:35,240 And actually, I'll visit that now. 450 00:40:35,240 --> 00:40:38,520 So essentially, that's worrisome. 451 00:40:38,520 --> 00:40:48,480 So in the Geeks for Geeks article, maybe I'll pin it again. 452 00:40:48,480 --> 00:40:58,240 They start to talk about essentially what do you want to optimize for in your logistic 453 00:40:58,240 --> 00:40:59,280 regression? 454 00:40:59,280 --> 00:41:07,960 So are you trying to get things precise, or are you trying to make sure that you don't 455 00:41:07,960 --> 00:41:09,400 miss any? 456 00:41:09,400 --> 00:41:19,240 And so from the laboratory's point of view, you don't want to be predicting a lot of passes 457 00:41:19,240 --> 00:41:22,140 that are actually failures. 458 00:41:22,140 --> 00:41:34,200 So in my book, it would be like this is the number that you would like to minimize. 459 00:41:34,200 --> 00:41:41,960 I think it's OK if you predict failures and there are actually passes. 460 00:41:41,960 --> 00:41:51,120 So I think this quadrant is not the worst in the world from the laboratory's point of 461 00:41:51,120 --> 00:41:52,800 view. 462 00:41:52,800 --> 00:41:56,760 But these are basic. 463 00:41:56,760 --> 00:42:00,200 Anything in this quadrant is essentially a missed opportunity. 464 00:42:00,200 --> 00:42:08,680 And the reason I say missed opportunity is so remember, so there's 750 samples here. 465 00:42:08,680 --> 00:42:16,080 So it's all about the value of time. 466 00:42:16,080 --> 00:42:26,240 So it takes about 10 minutes, give or take, to analyze one sample of hemp. 467 00:42:26,240 --> 00:42:40,400 So that's 125 hours to run 750 samples. 468 00:42:40,400 --> 00:42:47,400 And there'll probably be pauses in between when an analyst has to load on some more vials, 469 00:42:47,400 --> 00:42:57,960 because typically an instrument may only hold 50 to 100 vials that are being analyzed. 470 00:42:57,960 --> 00:43:05,640 So it may seem like a short amount of time, but it's a long amount of time. 471 00:43:05,640 --> 00:43:10,720 And there are a couple of laboratories doing these tests, but essentially you're trying 472 00:43:10,720 --> 00:43:17,920 to identify the failures as soon as possible, because the sooner you can identify the failure, 473 00:43:17,920 --> 00:43:28,120 you can notify the cultivator, and then they can begin their remedy process. 474 00:43:28,120 --> 00:43:35,420 So that's almost five days. 475 00:43:35,420 --> 00:43:46,760 So the idea is if you can load the failures onto the instrument first, then you can notify 476 00:43:46,760 --> 00:43:55,400 those cultivators almost five days sooner than you could otherwise. 477 00:43:55,400 --> 00:44:02,280 And that could mean a lot of money. 478 00:44:02,280 --> 00:44:09,480 So basically, let's say they just tested one field and it's high for THC. 479 00:44:09,480 --> 00:44:15,620 Well, maybe they have some other fields and they need to harvest those like right this 480 00:44:15,620 --> 00:44:19,200 second before it's too late. 481 00:44:19,200 --> 00:44:27,360 And so say if they waited five days, well now all of their fields may be at high THC. 482 00:44:27,360 --> 00:44:33,760 So it may seem like trivial or small or inconsequential. 483 00:44:33,760 --> 00:44:42,400 However, from my book, basically identifying these failures as soon as possible could just 484 00:44:42,400 --> 00:44:46,600 generate enormous savings later on. 485 00:44:46,600 --> 00:44:56,000 And it's just one of those things where just these small little ways that you do things. 486 00:44:56,000 --> 00:45:02,800 So the small little ways that you do analysis can have these enormous consequences. 487 00:45:02,800 --> 00:45:17,400 So even if this model may not look the best, it's only predicting 16 failures out of 750 488 00:45:17,400 --> 00:45:19,680 or so. 489 00:45:19,680 --> 00:45:27,800 Those are still 16 failures that you can load on the instrument first and just identify 490 00:45:27,800 --> 00:45:30,200 them that much sooner. 491 00:45:30,200 --> 00:45:39,480 And so long story short is what I would try to do is essentially build logit models to 492 00:45:39,480 --> 00:45:48,380 try to, of course, you want to maximize these two quadrants where you're predicting failures 493 00:45:48,380 --> 00:45:51,360 correctly and predicting passes correctly. 494 00:45:51,360 --> 00:46:02,440 However, I would continue to work on the model to try to reduce the number in this third 495 00:46:02,440 --> 00:46:09,280 quadrant and increase the number here or in the second and fourth quadrants. 496 00:46:09,280 --> 00:46:15,320 So that's how I would approach this analysis. 497 00:46:15,320 --> 00:46:18,440 So Keegan, question for you, make sure I'm tracking to this. 498 00:46:18,440 --> 00:46:24,200 So essentially by trying to predict which of the samples are more likely to fail, they 499 00:46:24,200 --> 00:46:28,240 kind of get bumped to the front of the line whenever they come into the lab. 500 00:46:28,240 --> 00:46:32,880 And by doing that and giving them first priority, you're giving that customer potentially more 501 00:46:32,880 --> 00:46:35,720 time to react to their failure. 502 00:46:35,720 --> 00:46:37,360 Exactly. 503 00:46:37,360 --> 00:46:44,420 And so there may be better ways to do it because I was thinking once it's failed, it may be 504 00:46:44,420 --> 00:46:46,200 too late. 505 00:46:46,200 --> 00:46:51,880 So it may be better to try to find things that are right at the brink. 506 00:46:51,880 --> 00:46:56,120 So right at like 0.28, 0.29. 507 00:46:56,120 --> 00:46:59,660 I don't know how you would go about doing that. 508 00:46:59,660 --> 00:47:12,280 So from the lab's point of view, it's always sort of a delicate, it's always delicate to 509 00:47:12,280 --> 00:47:19,880 just tell a cultivator, you've actually failed quality assurance testing or you've failed 510 00:47:19,880 --> 00:47:20,880 hemp testing. 511 00:47:20,880 --> 00:47:23,480 And it's just part of the business. 512 00:47:23,480 --> 00:47:28,800 It's just necessary. 513 00:47:28,800 --> 00:47:30,800 So it's often just a little bit delicate. 514 00:47:30,800 --> 00:47:35,560 So the sooner they know, they can go ahead and let them know the news. 515 00:47:35,560 --> 00:47:38,720 Like you've got high THC in your crop. 516 00:47:38,720 --> 00:47:47,080 And like I said, it's just that much sooner that they can begin essentially the remediation 517 00:47:47,080 --> 00:47:50,960 process for their next crop or adjacent crops. 518 00:47:50,960 --> 00:47:58,600 So this is something that, like I said, the hemp producers, they are, it is a problem 519 00:47:58,600 --> 00:47:59,600 for them at the moment. 520 00:47:59,600 --> 00:48:03,960 So there's a non-negligible portion failing and it's something that people are trying 521 00:48:03,960 --> 00:48:05,460 to pin down. 522 00:48:05,460 --> 00:48:13,040 And so from my point of view, even if their crop fails, but they have five additional 523 00:48:13,040 --> 00:48:20,640 days to try to work on it for next year, something's better than nothing. 524 00:48:20,640 --> 00:48:26,800 So I think that could go a long way. 525 00:48:26,800 --> 00:48:35,360 So that's essentially where I see the gains being from the laboratory point of view, from 526 00:48:35,360 --> 00:48:41,400 the cultivators point of view, these statistics may be a bit more interesting. 527 00:48:41,400 --> 00:48:50,360 So seeing, okay, we may want to, right, because different cultivators have a different risk 528 00:48:50,360 --> 00:48:55,080 preference, right, risk tolerance. 529 00:48:55,080 --> 00:48:58,800 So this is essentially your probability of failing. 530 00:48:58,800 --> 00:49:05,040 So some cultivators may have a higher preference for risk and they may want to push it out 531 00:49:05,040 --> 00:49:09,200 a ways to try to, right, because they're maximizing their CBD. 532 00:49:09,200 --> 00:49:14,600 And let's say they get revenue proportional to the CBD that they produce, right? 533 00:49:14,600 --> 00:49:19,400 So it's just a cost benefit analysis. 534 00:49:19,400 --> 00:49:26,560 So the longer you wait, the higher your benefit is, your CBD, but the higher your cost is, 535 00:49:26,560 --> 00:49:29,840 which is your probability of failing. 536 00:49:29,840 --> 00:49:43,600 So depending on their risk preference, you would just, one could think, well, simplifying 537 00:49:43,600 --> 00:49:48,840 things, you know, if these were just the only variables at hand, then you could just make 538 00:49:48,840 --> 00:49:49,840 a trade off. 539 00:49:49,840 --> 00:49:54,840 Okay, you know, how long do I want to wait for the amount of CBD given my probability 540 00:49:54,840 --> 00:49:55,840 of failing? 541 00:49:55,840 --> 00:49:59,920 And so then it would just be a simple cost benefit analysis, and you could use that to 542 00:49:59,920 --> 00:50:05,280 just choose, I want to harvest on this date. 543 00:50:05,280 --> 00:50:09,000 In reality, were you going to say something? 544 00:50:09,000 --> 00:50:10,920 Oh, I'm sorry, Keegan, go ahead. 545 00:50:10,920 --> 00:50:14,520 Or I was just going to say, in reality, there's a lot more factors. 546 00:50:14,520 --> 00:50:17,000 But do you have a comment? 547 00:50:17,000 --> 00:50:24,160 I did regarding your CBD to THC correlation plot. 548 00:50:24,160 --> 00:50:28,480 There was one outlier that was very interesting, which is right there on the right hand side 549 00:50:28,480 --> 00:50:35,640 where they're budding right up against the 0.3% THC threshold, but they're maximizing 550 00:50:35,640 --> 00:50:37,320 their CBD. 551 00:50:37,320 --> 00:50:43,000 That guy right there, to me, when I see that, that to me is like suspect. 552 00:50:43,000 --> 00:50:48,480 They either cracked the nut or something else is going on. 553 00:50:48,480 --> 00:50:49,480 At the 0.29%? 554 00:50:49,480 --> 00:50:57,600 Yeah, and they were all the way up there to almost 15, 14 and a half CBD. 555 00:50:57,600 --> 00:50:59,600 Yes. 556 00:50:59,600 --> 00:51:02,640 And yeah, that's, you're right. 557 00:51:02,640 --> 00:51:07,600 So they could have, they may have just gotten lucky. 558 00:51:07,600 --> 00:51:12,720 I don't think there's anything shady going on because we've got some pretty reputable 559 00:51:12,720 --> 00:51:13,720 labs here. 560 00:51:13,720 --> 00:51:25,040 So one would like to think that it's just luck of the draw. 561 00:51:25,040 --> 00:51:29,920 Yeah, maybe I'm being too cynical. 562 00:51:29,920 --> 00:51:35,280 But like I said, I mean, I think it's just an outlier. 563 00:51:35,280 --> 00:51:36,800 Like you said, anything's possible. 564 00:51:36,800 --> 00:51:48,120 I think maybe if you saw like a string of observations all at 0.29% that may be suspect. 565 00:51:48,120 --> 00:52:00,760 Yeah, and also the point by 10% CBD, which is kind of has low THC. 566 00:52:00,760 --> 00:52:03,240 So if it's right over there, right there. 567 00:52:03,240 --> 00:52:06,600 Yeah, that guy's doing pretty well, right? 568 00:52:06,600 --> 00:52:08,520 He's got a very low risk threshold. 569 00:52:08,520 --> 00:52:14,680 You know, if we're using the terms that you kind of defined through your logic model, 570 00:52:14,680 --> 00:52:17,480 he's got low risk and kind of high reward there. 571 00:52:17,480 --> 00:52:20,440 So that's a pretty good performance. 572 00:52:20,440 --> 00:52:23,560 And this is actually essentially what cultivators are trying to figure out. 573 00:52:23,560 --> 00:52:27,100 And like I said, it just luck of the draw getting these right. 574 00:52:27,100 --> 00:52:36,240 So it's basically, you're trying to figure out these, the right, the right strain. 575 00:52:36,240 --> 00:52:44,640 So you're trying to figure out the perfect strains to grow. 576 00:52:44,640 --> 00:52:50,920 And so basically, they're still figuring it out. 577 00:52:50,920 --> 00:53:02,280 So for example, these, the front range biosciences, they may want to start breeding hybrid number 578 00:53:02,280 --> 00:53:03,280 five. 579 00:53:03,280 --> 00:53:04,280 Right. 580 00:53:04,280 --> 00:53:14,880 And so I think essentially, I think this gets back to sort of essentially what I'm starting 581 00:53:14,880 --> 00:53:24,640 to narrow down on is it's about select. 582 00:53:24,640 --> 00:53:29,720 There's a lot of factors going on, but a lot comes down to selecting the right strain. 583 00:53:29,720 --> 00:53:36,880 So and there's, you know, and so that's why it's in high demand. 584 00:53:36,880 --> 00:53:37,880 Right. 585 00:53:37,880 --> 00:53:43,040 So just because you have hemp seeds doesn't necessarily mean they're going to be profitable. 586 00:53:43,040 --> 00:53:49,000 So you know, a lot of people out there, they may not know one way or the other, and they 587 00:53:49,000 --> 00:53:56,920 just purchase a lot of, you know, berry blossom seeds or presidential seeds. 588 00:53:56,920 --> 00:54:01,780 And they just, they just got, you know, the wrong genetics. 589 00:54:01,780 --> 00:54:09,280 And so I think this is what hemp producers are starting to learn is, you know, the strain 590 00:54:09,280 --> 00:54:10,760 you're growing really matters. 591 00:54:10,760 --> 00:54:19,520 So I think that's, I think that's what I'm starting to get to is because here we're starting 592 00:54:19,520 --> 00:54:24,400 to see that, okay, when you sample, it starts to matter. 593 00:54:24,400 --> 00:54:30,480 And so I'm starting to think, okay, that may be dictated by which strain you're actually 594 00:54:30,480 --> 00:54:33,760 growing. 595 00:54:33,760 --> 00:54:41,680 Were there multiple, multiple incidents at the same strain just being harvested at different 596 00:54:41,680 --> 00:54:42,680 times? 597 00:54:42,680 --> 00:54:44,120 Oh, yes. 598 00:54:44,120 --> 00:54:58,120 And so there is room for strain analysis here, because as you see, we've got a lot of repeats. 599 00:54:58,120 --> 00:55:01,440 So here are some of like the popular ones. 600 00:55:01,440 --> 00:55:13,760 So like, it looks like, like this cherry wine just looks to be, you know, unbelievably popular. 601 00:55:13,760 --> 00:55:18,480 But you know, it may not really be, you know, the best choice. 602 00:55:18,480 --> 00:55:22,880 So, so, or maybe it's a fantastic choice. 603 00:55:22,880 --> 00:55:27,680 And so I think, I think there's room there to be done, because basically, let's say you're 604 00:55:27,680 --> 00:55:31,560 a new, a new hemp producer out there. 605 00:55:31,560 --> 00:55:37,520 Well, you know, you're, you've got your, you've invested all this in your cultivation equipment, 606 00:55:37,520 --> 00:55:44,320 and you're about to buy, say, 30,000 seeds to go seed acres. 607 00:55:44,320 --> 00:55:48,880 Well, you want to make sure you pick out the right variety. 608 00:55:48,880 --> 00:55:55,080 And I'm sure every seed producer out there that you call is going to say that, oh, yeah, 609 00:55:55,080 --> 00:55:58,040 their seeds are fantastic. 610 00:55:58,040 --> 00:56:06,280 So so I think that's, that's where I think our analysis comes in, is, you know, if you're 611 00:56:06,280 --> 00:56:15,600 a hemp producer, you can begin to start tailoring our analysis and see if you can't, you know, 612 00:56:15,600 --> 00:56:19,720 start picking the varieties that are right for you. 613 00:56:19,720 --> 00:56:28,320 So it's known that over time, THC increases in these in hemp. 614 00:56:28,320 --> 00:56:34,360 So and there are a lot of cases of farmers harvesting, and it being tested like a month 615 00:56:34,360 --> 00:56:39,880 or two later, it passed when they harvested, but it fails later. 616 00:56:39,880 --> 00:56:44,840 So there's sort of this race against the clock. 617 00:56:44,840 --> 00:56:50,800 And I'm not sure I guess I'm from this data, it looks like CDB also goes up over time. 618 00:56:50,800 --> 00:56:53,760 But it's kind of trying to balance the two out. 619 00:56:53,760 --> 00:57:00,360 Yeah, definitely a trade off like you're saying, Keung, like a cost function here. 620 00:57:00,360 --> 00:57:06,680 Exactly, and so I think I think there's room for deeper economic analysis. 621 00:57:06,680 --> 00:57:11,480 And I think I think what Paul, I think you, you're you're real good at noting these outliers 622 00:57:11,480 --> 00:57:16,880 and their significance, because I think essentially, that's what you're you're aiming for. 623 00:57:16,880 --> 00:57:25,720 Because if you can find a stream like this, then you don't have quite the the the careful 624 00:57:25,720 --> 00:57:27,920 dance that you have to dance, right? 625 00:57:27,920 --> 00:57:35,600 Because you're growing something right here that it's going to be just such a such a fine 626 00:57:35,600 --> 00:57:43,680 line between passing and failing that, you know, that that could pose a that, you know, 627 00:57:43,680 --> 00:57:49,400 the probability of failing could pose such a high cost on a cultivator that it wouldn't 628 00:57:49,400 --> 00:57:51,320 be worth it or that they could. 629 00:57:51,320 --> 00:57:55,980 It's not uncommon that I mean, they could end up as a loss, right? 630 00:57:55,980 --> 00:58:02,160 Because if you act, it's not unheard of to, you know, to grow a lot of hemp, and then 631 00:58:02,160 --> 00:58:04,520 it fail. 632 00:58:04,520 --> 00:58:06,760 And then you'll you're in the red. 633 00:58:06,760 --> 00:58:10,320 Yeah, yeah, very risky. 634 00:58:10,320 --> 00:58:17,360 In in in from firsthand experience, a lot of hemp producers don't understand that risk 635 00:58:17,360 --> 00:58:18,720 getting into it. 636 00:58:18,720 --> 00:58:25,980 They think that they're growing a hemp strain, and they just they're just there's not going 637 00:58:25,980 --> 00:58:28,240 to be much THC in it. 638 00:58:28,240 --> 00:58:33,920 But as we're seeing, you know, there's there is a positive core, it appears to be a positive 639 00:58:33,920 --> 00:58:35,480 correlation here. 640 00:58:35,480 --> 00:58:42,560 So it's just that some of them, it's just a weaker, you know, weaker correlation. 641 00:58:42,560 --> 00:58:46,760 So yeah, good, good work, Keegan. 642 00:58:46,760 --> 00:58:49,800 And Charles, that's good stuff. 643 00:58:49,800 --> 00:58:55,760 The data that was collected and consolidated, the gentleman that you mentioned at the beginning 644 00:58:55,760 --> 00:59:00,600 of the presentation, did he reach out to each of the states separately to get this information? 645 00:59:00,600 --> 00:59:09,520 Yes, so the Washington State data is just from Washington State. 646 00:59:09,520 --> 00:59:18,400 So so in fact, touching on that real quick, didn't leave Charles 10 minutes, but I'll 647 00:59:18,400 --> 00:59:21,160 give you 10 minutes at the start of next meetup. 648 00:59:21,160 --> 00:59:34,440 So essentially, what Paul is referring to here is in you can find this in in our prior 649 00:59:34,440 --> 00:59:45,880 in our earlier meetups, there's several links here, essentially to a Washington State database. 650 00:59:45,880 --> 00:59:52,040 And so this data was collected by Jim McCray, who's a researcher in Washington State and 651 00:59:52,040 --> 00:59:53,960 shared with me. 652 00:59:53,960 --> 01:00:00,120 And then Charles has done some analysis here, just trying to predict the probability of 653 01:00:00,120 --> 01:00:02,240 a sample failing. 654 01:00:02,240 --> 01:00:06,360 And so Charles, I apologize to you that I didn't leave you 10 minutes. 655 01:00:06,360 --> 01:00:15,880 And so if you want to go over this at the beginning of next week, we can set that aside. 656 01:00:15,880 --> 01:00:18,040 Okay, that's cool. 657 01:00:18,040 --> 01:00:19,120 All right. 658 01:00:19,120 --> 01:00:25,160 And then for the rest of the crew, you can look through Charles's work here, because 659 01:00:25,160 --> 01:00:32,680 he's done some fantastic work, essentially seeing if we can, similarly to what we've 660 01:00:32,680 --> 01:00:40,200 done today, predict the probability of failing, given what sample type it is. 661 01:00:40,200 --> 01:00:44,560 But I'll let Charles get more into that next week. 662 01:00:44,560 --> 01:00:47,000 All right. 663 01:00:47,000 --> 01:00:48,000 Looking forward to that one, Charles. 664 01:00:48,000 --> 01:00:49,000 That looks good. 665 01:00:49,000 --> 01:00:50,000 All right. 666 01:00:50,000 --> 01:00:56,680 Well, I'm going to end the presentation here. 667 01:00:56,680 --> 01:01:05,840 And so real quick, are there any quick comments, questions, concerns before we call it a day? 668 01:01:05,840 --> 01:01:07,240 Nope. 669 01:01:07,240 --> 01:01:11,840 Well, it's been awesome having everybody. 670 01:01:11,840 --> 01:01:13,800 It's great having some new faces here. 671 01:01:13,800 --> 01:01:19,620 So it's awesome seeing you, Gabrielle, Eri, Megan, and of course, Paul, Charles, and Heather. 672 01:01:19,620 --> 01:01:21,360 So it's been fun. 673 01:01:21,360 --> 01:01:28,200 So let's keep in touch, keep bouncing ideas off each other. 674 01:01:28,200 --> 01:01:33,760 And then next week, we'll start with Charles's analysis and keep doing analytics. 675 01:01:33,760 --> 01:01:35,760 Sounds good. 676 01:01:35,760 --> 01:01:39,440 All right, everyone. 677 01:01:39,440 --> 01:01:42,880 Have a productive week and until next week, keep being awesome. 678 01:01:42,880 --> 01:01:43,880 Bye now. 679 01:01:43,880 --> 01:01:44,880 Bye-bye. 680 01:01:44,880 --> 01:02:00,960 Bye.