1 00:00:03,930 --> 00:00:15,450 Okay, so in this, uh, chapter, we are going to, uh, work in a another basic rack application, 2 00:00:15,450 --> 00:00:24,000 but in this case with a little bit more sophisticated vector database, we are going to use the database 3 00:00:24,000 --> 00:00:28,860 Deep Lake, which is a database vector database. 4 00:00:28,860 --> 00:00:36,270 Uh, that is going to ask a credentials, uh, from us in order to be able to connect with it. 5 00:00:36,270 --> 00:00:42,210 This is the most common way to operate in a professional way with vector databases, as you will see. 6 00:00:42,240 --> 00:00:49,980 And this is a process that is going to be, uh, used whenever you use a professional vector database 7 00:00:49,980 --> 00:00:52,770 pinecone, V8, Redis, whatever. 8 00:00:52,770 --> 00:00:59,400 So the process you are going to see in the exercise is going to be very similar to the process you are 9 00:00:59,400 --> 00:01:07,230 going to, uh, perform whenever you have to work with a professional vector database. 10 00:01:08,270 --> 00:01:16,100 So the problem we are going to solve in this exercise is to test a drug application with a different 11 00:01:16,100 --> 00:01:19,640 vector database and access credentials. 12 00:01:19,640 --> 00:01:28,580 And later we are going to add to the external document A used by the RAC application, a data to the 13 00:01:28,580 --> 00:01:31,280 external document used by the RAC applications. 14 00:01:31,280 --> 00:01:33,470 Okay, so a couple of things. 15 00:01:33,470 --> 00:01:38,990 We are already familiar with a basic applications. 16 00:01:38,990 --> 00:01:40,310 We have done it before. 17 00:01:40,310 --> 00:01:48,770 In this case we are going to go a little bit, uh, further and we are going to uh connect with a little 18 00:01:48,770 --> 00:01:52,400 bit more sophisticated vector database in this case. 19 00:01:52,670 --> 00:01:53,600 So. 20 00:01:54,440 --> 00:01:58,430 As I was saying, we are going to use a deep Lake vector database. 21 00:01:58,430 --> 00:02:04,010 Deep Lake is a database owned by the company Active Loop. 22 00:02:04,010 --> 00:02:06,440 This is how you write Active Loop. 23 00:02:06,440 --> 00:02:07,940 So you go to their website. 24 00:02:07,940 --> 00:02:16,550 You will see how you can create a free account there, how you can get, you know, your token in order 25 00:02:16,550 --> 00:02:24,680 to start using the database and how you can get the ID a the organization ID that is a requirement for 26 00:02:24,680 --> 00:02:26,630 you to start using the databases, etc.. 27 00:02:26,630 --> 00:02:35,210 So all this you have to do in their website whatever question doubt you have regarding active loop or 28 00:02:35,210 --> 00:02:39,080 vector database or whatever operations we are doing here. 29 00:02:39,080 --> 00:02:47,960 Remember the usual places you can find answers chat, GPT, Google, StackOverflow, etc. etc.. 30 00:02:47,960 --> 00:02:52,910 Okay, so this is the process we are going to follow in this exercise. 31 00:02:52,910 --> 00:02:53,750 At the beginning. 32 00:02:53,750 --> 00:03:01,880 As always, we uh, load the credentials in order to use the OpenAI API. 33 00:03:01,880 --> 00:03:02,480 Okay. 34 00:03:02,480 --> 00:03:07,280 So as you remember we have this credentials in the dot env file. 35 00:03:07,310 --> 00:03:10,280 Do you remember this was explained chapter seven. 36 00:03:10,430 --> 00:03:15,680 So the first thing we are going to do is to load the deep lake credentials. 37 00:03:15,680 --> 00:03:21,680 So we have our deep lake credentials in our dot m file. 38 00:03:21,680 --> 00:03:22,220 Okay. 39 00:03:22,220 --> 00:03:27,260 This is a confidential file that you have already created probably. 40 00:03:27,260 --> 00:03:32,150 And uh, this is what, uh, every uh, software engineer uses. 41 00:03:32,150 --> 00:03:39,200 Whenever we work with applications, we, we create a dot env file, and there we store all kind of 42 00:03:39,200 --> 00:03:40,850 passwords, credentials, etc.. 43 00:03:40,850 --> 00:03:51,350 So what we are seeing here is, uh, getting the active look token and the active look or ID from the 44 00:03:51,350 --> 00:03:52,850 dot m file. 45 00:03:53,120 --> 00:04:01,700 The next thing we do is we, uh, name the new database, uh, we are going to create. 46 00:04:01,700 --> 00:04:09,680 So in this case, uh, this is the name we are giving to the data set we are creating. 47 00:04:09,950 --> 00:04:13,310 And, uh, we create, we load the dependencies. 48 00:04:13,310 --> 00:04:20,750 In this case, you are already familiar with OpenAI embeddings with recursive characters, uh, text 49 00:04:20,750 --> 00:04:27,710 splitter and with retrieval QA because this, uh, we have already used in previous exercises whenever 50 00:04:27,710 --> 00:04:30,350 we have created a basic application. 51 00:04:30,350 --> 00:04:40,760 But in this case we are using a new module called Deep Lake in order to connect long chain with this 52 00:04:40,760 --> 00:04:41,510 vector store. 53 00:04:41,510 --> 00:04:49,100 So as easy as that, once we have these dependencies loaded or these modules or plugins, whatever you 54 00:04:49,100 --> 00:04:57,020 want to call them, what we do is, uh, pay attention to the external knowledge document we are going 55 00:04:57,020 --> 00:04:57,440 to use. 56 00:04:57,440 --> 00:05:08,390 So remember, a rack application is an application that provides the LM foundation model with an additional 57 00:05:08,390 --> 00:05:09,530 information. 58 00:05:09,530 --> 00:05:17,870 So as you know, ChatGPT for example, is trained with all the information and that is on the internet 59 00:05:17,870 --> 00:05:25,070 or more or less all of all of it, but it doesn't know anything about your company, your book, your 60 00:05:25,070 --> 00:05:26,570 user guide, whatever. 61 00:05:26,570 --> 00:05:34,010 All the private documents you have are not included in the knowledge that ChatGPT has right now. 62 00:05:34,010 --> 00:05:38,540 And the way we can provide this knowledge and we can start interacting. 63 00:05:38,540 --> 00:05:45,740 Interacting with this knowledge via a LM application is uh, with a rack application, right? 64 00:05:45,740 --> 00:05:51,290 So the first thing we do is we need to, uh, uh, we need to. 65 00:05:52,100 --> 00:05:56,900 Tell the application what is our where is our private document. 66 00:05:57,380 --> 00:06:06,800 So you can use a PDF as a private document or a text file or whatever word document, whatever kind 67 00:06:06,800 --> 00:06:07,490 of file. 68 00:06:07,700 --> 00:06:12,320 In this case, we are going to use a very simple document. 69 00:06:12,320 --> 00:06:18,740 We are going to create a list of only two, uh, items. 70 00:06:18,890 --> 00:06:27,410 The first item we are going to talk about, one, uh, curious fact about United States in this case 71 00:06:27,410 --> 00:06:32,330 is, uh, the date where United States celebrates the Independence Day. 72 00:06:32,330 --> 00:06:40,310 And in the same chunk of information we tell them, however, the Declaration of Independence was passed 73 00:06:40,310 --> 00:06:43,010 on July the 2nd. 74 00:06:43,010 --> 00:06:43,670 Right. 75 00:06:43,670 --> 00:06:48,950 And it was only officially ratified on, uh, July the 4th. 76 00:06:48,950 --> 00:06:52,250 So this is the special information. 77 00:06:52,250 --> 00:06:53,360 We are included here. 78 00:06:53,360 --> 00:06:59,990 And this is something that we are going to be asking about when we test this application. 79 00:06:59,990 --> 00:07:06,920 The second piece of information we we include in this private document is okay. 80 00:07:06,920 --> 00:07:13,160 The first documented European to arrive in North America was the Spaniard Juan Ponce de Leon. 81 00:07:13,160 --> 00:07:18,470 Okay, so this is the external knowledge document we are going to use. 82 00:07:18,470 --> 00:07:21,860 In this case, we are creating right now the external document. 83 00:07:21,860 --> 00:07:27,980 But you could use, you know, a document that you have already in your company or, or personal use 84 00:07:27,980 --> 00:07:28,640 or whatever. 85 00:07:28,880 --> 00:07:37,640 So remember how we work in a large application, the first thing we do is to divide the document in 86 00:07:37,640 --> 00:07:45,530 smaller chunks of text in order to this is the document, and we are going to divide this document in 87 00:07:45,650 --> 00:07:47,840 smaller chunks of text. 88 00:07:47,960 --> 00:07:52,310 In order to do that, we use the text splitter. 89 00:07:52,310 --> 00:07:55,850 So the first thing we do is we configure the text splitter. 90 00:07:55,850 --> 00:07:59,930 And the second we create the document chunks. 91 00:08:00,020 --> 00:08:07,850 And if we print the length of this document chunks we will see that we have two chunks. 92 00:08:07,850 --> 00:08:15,170 So we have divided this document into different chunks or fragments of text. 93 00:08:15,800 --> 00:08:24,290 As you remember, the next operation we have to, uh, perform in a drag application is to transform 94 00:08:24,290 --> 00:08:30,320 this chunks of text into numbers, what we call the embeddings. 95 00:08:30,320 --> 00:08:34,100 Because as you remember, the computer works. 96 00:08:34,750 --> 00:08:38,650 Much better with numbers than with letters, right? 97 00:08:38,650 --> 00:08:42,820 So we convert our chunks of text into embeddings. 98 00:08:42,820 --> 00:08:44,440 How do we do it? 99 00:08:44,470 --> 00:08:47,530 Same thing we did in the previous exercise. 100 00:08:47,530 --> 00:08:55,570 We are going to use the OpenAI embeddings module in order to build these embeddings, and we are going 101 00:08:55,570 --> 00:09:01,480 to use it in the a that database we have created. 102 00:09:01,480 --> 00:09:12,100 So using the Deep Lake module we are going to tell uh, our application where is our uh database and 103 00:09:12,100 --> 00:09:20,710 what uh, way are we going to use in order to transform the chunks of text into a number. 104 00:09:20,710 --> 00:09:23,440 So how are we going to build the embeddings? 105 00:09:23,440 --> 00:09:24,070 Okay. 106 00:09:24,070 --> 00:09:32,260 So in this case what you are actually doing is to create or, you know, start including data in your 107 00:09:32,260 --> 00:09:33,850 vector database. 108 00:09:33,850 --> 00:09:43,360 Since this is not the first time I execute this command, the the notebook is telling me, okay, this 109 00:09:43,360 --> 00:09:45,640 database already exists. 110 00:09:45,730 --> 00:09:48,070 Uh, it's already there. 111 00:09:48,070 --> 00:09:53,170 If you are creating a new one, you are going to have a different message here telling you, okay, 112 00:09:53,170 --> 00:10:00,880 your database has been created or whatever, pay attention because here we are including in the pip 113 00:10:00,880 --> 00:10:03,220 install the Blake enterprise. 114 00:10:03,220 --> 00:10:11,380 So this is a command that we will need in order to execute this command okay. 115 00:10:11,380 --> 00:10:19,420 If you don't have the Blake Enterprise installed in your virtual environment you will need to do that. 116 00:10:19,420 --> 00:10:20,350 You can do it. 117 00:10:20,350 --> 00:10:24,970 You can do you can install this module from the Jupyter notebook. 118 00:10:24,970 --> 00:10:28,720 And in order to do that you have to write this one. 119 00:10:28,720 --> 00:10:36,250 What I have here is this commented because since I already have it, uh, install and I don't want to 120 00:10:36,250 --> 00:10:38,560 install it again, it's not necessary for me. 121 00:10:38,560 --> 00:10:46,120 But in your case, if it is the first time you install it, you can do this and then, uh, you know, 122 00:10:46,120 --> 00:10:54,160 uh, caps and enter and you have a, you have that installed or you can do it, uh, via the terminal. 123 00:10:54,160 --> 00:11:00,100 So you can write pip install, uh, Deep Lake Enterprise in the terminal. 124 00:11:00,100 --> 00:11:03,250 And it will be installed in the virtual environment as well. 125 00:11:03,250 --> 00:11:11,470 Remember that we use this exclamation mark in order to tell the notebook that we are going to do a terminal 126 00:11:11,470 --> 00:11:14,440 operation from the notebook. 127 00:11:14,440 --> 00:11:14,770 Okay. 128 00:11:14,770 --> 00:11:22,210 So here we are creating the well we are entering content in the database okay. 129 00:11:22,630 --> 00:11:23,350 With this. 130 00:11:23,350 --> 00:11:30,430 With this one we are going to load the chunks and transform that the chunks into embeddings okay. 131 00:11:30,430 --> 00:11:32,530 With this operation here. 132 00:11:32,530 --> 00:11:35,650 And once we have that ready we can start. 133 00:11:35,650 --> 00:11:37,900 You know testing our application. 134 00:11:37,900 --> 00:11:39,760 It's as easy as that. 135 00:11:40,370 --> 00:11:48,770 So we create our LM instance and we create our retrieval QA chain. 136 00:11:49,040 --> 00:11:57,800 Uh, so we we give the, the the model we are using the chain type stuff as you know, is the most, 137 00:11:57,800 --> 00:12:03,890 uh, popular, the most used type of chain we are going to use for retrieval QA. 138 00:12:03,920 --> 00:12:09,680 You have a couple of alternatives here, but usually staff is the most common. 139 00:12:09,680 --> 00:12:15,650 You can go to the documentation to look about the other 2 or 3 you have. 140 00:12:15,650 --> 00:12:22,070 And then we have to tell, uh, how are we going to retrieve the information in this case we enter the 141 00:12:22,070 --> 00:12:27,050 database, we have created the function as retriever and we are ready to go. 142 00:12:27,470 --> 00:12:34,700 In order to start asking questions to our private document, we run the new chain. 143 00:12:35,180 --> 00:12:43,490 And in this case, as you can see, we are asking, uh, the, the, the application about, uh, you 144 00:12:43,490 --> 00:12:50,960 know, the actual date when the Declaration of Independence in United States was actually passed, as 145 00:12:50,960 --> 00:12:55,160 we know it was July the 2nd, not July the 4th. 146 00:12:55,640 --> 00:13:06,470 Uh, so then we have here an additional, uh, operation, how to add data to a vector database that 147 00:13:06,470 --> 00:13:07,850 we have already created. 148 00:13:07,850 --> 00:13:14,810 So let's see that you have new information you want to add to your private knowledge document. 149 00:13:14,810 --> 00:13:15,230 Right. 150 00:13:15,230 --> 00:13:17,360 So this is the way you can do that. 151 00:13:17,810 --> 00:13:21,410 So this is the additional information we can include to our document. 152 00:13:21,410 --> 00:13:28,790 And what we are going to do is we are going to add this document directly to our vector database. 153 00:13:28,790 --> 00:13:35,480 In order to do that, the first thing we do is we use the text splitter to divide this information into 154 00:13:35,480 --> 00:13:36,500 text chunks. 155 00:13:36,500 --> 00:13:44,540 And second, we use the add documents function in order to transform the text chunks into numbers into 156 00:13:44,540 --> 00:13:45,230 embeddings. 157 00:13:45,230 --> 00:13:45,890 Okay. 158 00:13:45,890 --> 00:13:54,860 So uh, once you have this, you can start asking questions about the new information you have added 159 00:13:54,860 --> 00:13:56,420 to the vector database. 160 00:13:56,420 --> 00:14:00,560 In this case, we are asking what is the largest state in the US. 161 00:14:00,560 --> 00:14:02,990 And here we are saying that this is Alaska. 162 00:14:02,990 --> 00:14:10,460 And in the second piece of uh, knowledge we enter here, what we are saying is that in some states 163 00:14:10,460 --> 00:14:17,660 and in some cities, big cities of the United States, you have one special style of pizza, right? 164 00:14:17,660 --> 00:14:19,880 So we are giving this information here. 165 00:14:19,880 --> 00:14:26,120 And in the last question we are asking, tell me three states with their own style of pizza. 166 00:14:26,120 --> 00:14:33,170 In this case, you can see that the application is giving us the right answer, but just in part because 167 00:14:33,170 --> 00:14:38,000 it is giving us a states that have a particular pizza style. 168 00:14:38,000 --> 00:14:47,000 But in this case, it is, uh, giving us five, uh, states, not three as we, uh, as we asked. 169 00:14:47,000 --> 00:14:56,090 So we see that we will, uh, face, you know, this kind of, uh, not 100% accurate responses from 170 00:14:56,450 --> 00:14:57,350 applications. 171 00:14:57,350 --> 00:15:01,610 And we will learn ways to make them more and more accurate. 172 00:15:01,610 --> 00:15:02,180 Okay. 173 00:15:02,180 --> 00:15:11,390 So in this chapter we have seen how to create a basic lag application with a little bit more sophisticated 174 00:15:11,390 --> 00:15:12,530 database.