1 00:00:00,000 --> 00:00:03,000 So finally guys, you can see the training has got completed. 2 00:00:03,000 --> 00:00:05,000 It took somewhere around eight minutes. 3 00:00:05,000 --> 00:00:07,000 Uh, and 8.8 second. 4 00:00:07,000 --> 00:00:08,000 Okay. 5 00:00:08,000 --> 00:00:13,000 And, uh, these are the number of epochs I was able to get a maximum accuracy. 6 00:00:14,000 --> 00:00:16,000 The training accuracy was somewhere around 94%. 7 00:00:16,000 --> 00:00:20,000 And the validation accuracy was point A, that is 81%. 8 00:00:20,000 --> 00:00:25,000 If I probably go ahead and do another ten more epochs, I feel that still the accuracy will keep on 9 00:00:25,000 --> 00:00:25,000 increasing. 10 00:00:25,000 --> 00:00:30,000 Okay, but I just wanted to show you till ten epochs so that, uh, you know, it won't take more time 11 00:00:30,000 --> 00:00:31,000 for the training. 12 00:00:31,000 --> 00:00:33,000 But initially what I thought, you know. 13 00:00:33,000 --> 00:00:35,000 So over here, please focus on this. 14 00:00:35,000 --> 00:00:35,000 Right. 15 00:00:35,000 --> 00:00:36,000 So loss and accuracy. 16 00:00:36,000 --> 00:00:39,000 Initially loss was shown as a very high value. 17 00:00:39,000 --> 00:00:40,000 I got worried okay. 18 00:00:40,000 --> 00:00:42,000 And then accuracy increased. 19 00:00:42,000 --> 00:00:45,000 Then you can see loss also increased. 20 00:00:45,000 --> 00:00:49,000 I think this is a kind of a glitch that has actually happened because after this you will be able to 21 00:00:49,000 --> 00:00:52,000 see the loss was in decimals in itself. 22 00:00:52,000 --> 00:00:54,000 And this was also getting decreased. 23 00:00:54,000 --> 00:00:54,000 Okay. 24 00:00:54,000 --> 00:00:59,000 But if you see with respect to validation loss here there was no glitches as such. 25 00:00:59,000 --> 00:01:04,000 So from 0.59 we came back to 0.50 which is good enough okay. 26 00:01:04,000 --> 00:01:08,000 And we are able to achieve the validation accuracy of 81%, which is good enough. 27 00:01:08,000 --> 00:01:15,000 Uh, now what I am actually going to do is that I'm just going to go ahead and save this entire file. 28 00:01:15,000 --> 00:01:22,000 Now, in order to save this, uh, I will just go ahead and say, hey, um, model dot, uh, or whatever 29 00:01:22,000 --> 00:01:28,000 models we specifically have, we can just go ahead and save into save it into the H5 file. 30 00:01:28,000 --> 00:01:28,000 Right. 31 00:01:28,000 --> 00:01:33,000 So here I will just go ahead and save my model file. 32 00:01:33,000 --> 00:01:33,000 Okay. 33 00:01:33,000 --> 00:01:35,000 I'll give you an assignment. 34 00:01:35,000 --> 00:01:39,000 You can also keep all these logs in your log folder and use TensorBoard to probably display how we did 35 00:01:39,000 --> 00:01:40,000 it in the end. 36 00:01:40,000 --> 00:01:44,000 Okay so here I'm just going to write model dot save. 37 00:01:44,000 --> 00:01:50,000 Let me just go ahead and write simple underscore RNN underscore IMDb. 38 00:01:51,000 --> 00:01:55,000 IMDb dot uh dot h5 file okay. 39 00:01:56,000 --> 00:01:56,000 So once I execute it. 40 00:01:56,000 --> 00:02:01,000 So this particular file may have got saved over here okay. 41 00:02:01,000 --> 00:02:03,000 Now, quickly, uh, let's do one thing. 42 00:02:03,000 --> 00:02:07,000 Let's also create, uh, one prediction dot ipynb file. 43 00:02:08,000 --> 00:02:12,000 Uh, the reason is very simple because I also need to create a Streamlit app right later on. 44 00:02:12,000 --> 00:02:16,000 So how to properly load this entire model will be discussing about that. 45 00:02:16,000 --> 00:02:18,000 Now this will be very simple. 46 00:02:18,000 --> 00:02:23,000 First of all, uh, I will go ahead and import all this particular libraries. 47 00:02:23,000 --> 00:02:27,000 So let's go ahead and import this data sets pre-processing all these things. 48 00:02:28,000 --> 00:02:33,000 Now from this particular IMDb data sets let's get all the word index okay. 49 00:02:33,000 --> 00:02:34,000 Load the word index. 50 00:02:34,000 --> 00:02:37,000 And we also have this reverse word index okay. 51 00:02:37,000 --> 00:02:39,000 This also I have already explained you. 52 00:02:40,000 --> 00:02:44,000 Then uh, we will go ahead and load our H5 file. 53 00:02:45,000 --> 00:02:45,000 Okay. 54 00:02:46,000 --> 00:02:47,000 Model dot h l file file. 55 00:02:47,000 --> 00:02:51,000 If you go ahead and see the model dot summary okay. 56 00:02:51,000 --> 00:02:53,000 Model dot summary, you'll be able to see it. 57 00:02:53,000 --> 00:02:55,000 One more thing that I forgot to show you. 58 00:02:55,000 --> 00:02:56,000 Right. 59 00:02:56,000 --> 00:03:03,000 Uh, that is, uh, how do you, uh, how do you probably see model dot summary? 60 00:03:03,000 --> 00:03:04,000 If you. 61 00:03:04,000 --> 00:03:08,000 Right, you'll be able to see the model summary if I also want to probably get the weights. 62 00:03:08,000 --> 00:03:10,000 And I hope I've never shown you this. 63 00:03:10,000 --> 00:03:12,000 So here you can actually see all the configuration. 64 00:03:13,000 --> 00:03:15,000 If you want to probably go ahead and see get all the weights. 65 00:03:15,000 --> 00:03:18,000 So here are all the weights that is used with respect to the parameter. 66 00:03:18,000 --> 00:03:20,000 So this is also very handy. 67 00:03:20,000 --> 00:03:22,000 Sometime we don't even save the file that five file. 68 00:03:22,000 --> 00:03:26,000 What we do is that we save the architecture, we save the weights and then we try to combine it. 69 00:03:26,000 --> 00:03:27,000 Okay. 70 00:03:27,000 --> 00:03:28,000 So that way also we can do. 71 00:03:28,000 --> 00:03:30,000 But right now let's move. 72 00:03:30,000 --> 00:03:31,000 Focus on this H5 file. 73 00:03:31,000 --> 00:03:35,000 Later on when I get an opportunity I will also show you that okay. 74 00:03:36,000 --> 00:03:39,000 Then uh, we require two helper function. 75 00:03:39,000 --> 00:03:42,000 One is for decoding the review okay. 76 00:03:42,000 --> 00:03:44,000 And one is for pre-process input. 77 00:03:44,000 --> 00:03:50,000 Now in pre-process input what we do we basically do the padding right sequence dot pad sequences. 78 00:03:50,000 --> 00:03:57,000 And here the max length is 500 I hope in simple RNN what is the max length that we took. 79 00:03:57,000 --> 00:03:58,000 Right. 80 00:03:58,000 --> 00:04:00,000 I guess it was somewhere around 500. 81 00:04:00,000 --> 00:04:02,000 Yeah I am sure 500 only. 82 00:04:02,000 --> 00:04:06,000 Okay, so here, uh, this two will be my utility function. 83 00:04:06,000 --> 00:04:07,000 Okay. 84 00:04:07,000 --> 00:04:10,000 One is for decoding the review and one is for pre-process text. 85 00:04:10,000 --> 00:04:15,000 Uh, whenever the user input gives, uh, gives an user input, we have to probably take that particular 86 00:04:15,000 --> 00:04:15,000 word. 87 00:04:15,000 --> 00:04:18,000 We have to encode that entire review, we have to pad that review. 88 00:04:18,000 --> 00:04:20,000 And finally we get that padded review. 89 00:04:20,000 --> 00:04:25,000 Then only we'll be able to pass it right, uh, to our uh, model. 90 00:04:25,000 --> 00:04:27,000 Now this is very important. 91 00:04:27,000 --> 00:04:31,000 We will go ahead and create our prediction function okay. 92 00:04:31,000 --> 00:04:33,000 Prediction function. 93 00:04:33,000 --> 00:04:37,000 Now in this prediction function we you know that we have loaded the model, right. 94 00:04:37,000 --> 00:04:43,000 So I will just go ahead and write definition predict underscore sentiment. 95 00:04:44,000 --> 00:04:48,000 And here I'm just going to see the review. 96 00:04:48,000 --> 00:04:53,000 And inside this review what I will go ahead and do is that first of all, whenever I get the review 97 00:04:53,000 --> 00:04:59,000 I have to probably pass, uh, through this pre-process underscore text. 98 00:04:59,000 --> 00:05:00,000 Right. 99 00:05:00,000 --> 00:05:02,000 And I will give the review over here right Right. 100 00:05:02,000 --> 00:05:06,000 Once I give the I call this function what is going to happen. 101 00:05:06,000 --> 00:05:08,000 All the padding is basically going to get applied. 102 00:05:08,000 --> 00:05:12,000 First of all, I'm going to get the encoded format for that particular review. 103 00:05:12,000 --> 00:05:16,000 And then we are going to pad all the sequence uh pre padding. 104 00:05:16,000 --> 00:05:20,000 Uh, so I'm going to probably get that zeros and ones uh that I want. 105 00:05:20,000 --> 00:05:26,000 Then I will go ahead and save this particular variable called as preprocessed underscore input over 106 00:05:26,000 --> 00:05:26,000 here. 107 00:05:26,000 --> 00:05:32,000 Okay Then once we get this I'm going to use model dot predict. 108 00:05:33,000 --> 00:05:36,000 And I'm going to pass my preprocessed input over here. 109 00:05:37,000 --> 00:05:37,000 Right. 110 00:05:37,000 --> 00:05:42,000 Once I pass this over here I will be able to make my prediction. 111 00:05:43,000 --> 00:05:50,000 So once I'm able to make the prediction I will quickly go ahead and write one very important condition. 112 00:05:50,000 --> 00:05:53,000 So I'll say, hey, uh, I'll create a variable sentiment. 113 00:05:53,000 --> 00:05:54,000 I'll say, hey, positive. 114 00:05:54,000 --> 00:05:57,000 If prediction of zero zero is greater than 0.5, else negative. 115 00:05:57,000 --> 00:05:57,000 Okay. 116 00:05:57,000 --> 00:06:02,000 And then we are going to return both sentiment and prediction whatever values we are actually able to 117 00:06:02,000 --> 00:06:02,000 get. 118 00:06:02,000 --> 00:06:07,000 So this will basically be my function which is called as predict uh sentiment okay. 119 00:06:07,000 --> 00:06:11,000 Now let's go ahead and see with respect to some examples. 120 00:06:12,000 --> 00:06:16,000 So this is my next step here I'm taking uh review. 121 00:06:16,000 --> 00:06:20,000 The movie was fantastic, the acting was great, and the plot was thrilling. 122 00:06:20,000 --> 00:06:22,000 Okay, so this is my review. 123 00:06:22,000 --> 00:06:23,000 Okay, now see this? 124 00:06:23,000 --> 00:06:29,000 Uh, as soon as I go ahead and pass my predict sentiment and I pass my example review over here, I'll 125 00:06:29,000 --> 00:06:30,000 be getting two values. 126 00:06:30,000 --> 00:06:35,000 One is my sentiment comma prediction right now. 127 00:06:35,000 --> 00:06:38,000 Or I can just go ahead and write my score. 128 00:06:38,000 --> 00:06:39,000 Okay. 129 00:06:39,000 --> 00:06:40,000 Sentiment and score. 130 00:06:41,000 --> 00:06:45,000 Now I will go ahead and make sure that print with the F string. 131 00:06:45,000 --> 00:06:52,000 And here you will be able to see this okay so I've got my example review sentiment prediction score. 132 00:06:52,000 --> 00:06:55,000 If I go ahead and execute it here you can see the movie was fantastic. 133 00:06:55,000 --> 00:06:55,000 The acting was great. 134 00:06:55,000 --> 00:06:57,000 Sentiment is positive. 135 00:06:57,000 --> 00:06:59,000 The prediction score is 0.811. 136 00:06:59,000 --> 00:07:05,000 So this prediction of zero zero when we are using in short, we are getting the score right and this 137 00:07:05,000 --> 00:07:05,000 is greater than 0.5. 138 00:07:05,000 --> 00:07:07,000 So that is the reason we are getting positive. 139 00:07:08,000 --> 00:07:12,000 So in this video, uh, we have saved the, uh, RNN file. 140 00:07:12,000 --> 00:07:14,000 We have loaded it over here. 141 00:07:14,000 --> 00:07:17,000 We got a very good accuracy with respect to the simple RNN. 142 00:07:17,000 --> 00:07:17,000 Right. 143 00:07:17,000 --> 00:07:20,000 And we also able to make the prediction. 144 00:07:20,000 --> 00:07:22,000 Now the same thing will try to create a Streamlit app. 145 00:07:22,000 --> 00:07:25,000 And then we'll create our entire web app. 146 00:07:25,000 --> 00:07:30,000 And finally we'll also go ahead and deploy the our Streamlit web app in the Streamlit cloud that I will 147 00:07:30,000 --> 00:07:32,000 be showing you in the next video. 148 00:07:32,000 --> 00:07:33,000 Thank you.