1
00:00:00,660 --> 00:00:07,510
In the last video we saw how we can save our existing trade model to a file and then load it back in

2
00:00:08,260 --> 00:00:13,210
and then make some predictions that get the exact same results as the original model that we save to

3
00:00:13,210 --> 00:00:14,070
file.

4
00:00:14,110 --> 00:00:16,480
We used pickle to do that in this video.

5
00:00:16,470 --> 00:00:20,530
We're going to see how we can do the same thing with the job live module.

6
00:00:21,010 --> 00:00:25,690
And again if you don't know something you can always go what is job lib

7
00:00:28,740 --> 00:00:33,720
job live running Python functions as pipeline jobs and you could go through the documentation here to

8
00:00:33,720 --> 00:00:34,660
see how you doing.

9
00:00:34,740 --> 00:00:40,080
But let's just see an example of how we do the exact same thing here with the job live module to create

10
00:00:40,080 --> 00:00:47,470
a heading go job lib change that to mark down hit shift and enter.

11
00:00:47,510 --> 00:00:54,080
So the first thing that we going to do we can go from job lead import dump and load.

12
00:00:54,150 --> 00:00:57,150
You might be out a guess what these two functions do.

13
00:00:57,150 --> 00:01:05,190
So to save model to file we'll go dump GSL F because we're gonna say the exact same model that we did

14
00:01:05,190 --> 00:01:05,860
with pickle.

15
00:01:05,910 --> 00:01:15,270
So our grid search classifier we're gonna pass it file name we'll go G.S. random forest model one dot

16
00:01:15,450 --> 00:01:16,290
job lib.

17
00:01:17,070 --> 00:01:18,410
So that's the difference here.

18
00:01:18,570 --> 00:01:20,210
We can positive file name.

19
00:01:20,280 --> 00:01:26,190
We have to press equals with pickle we used up scale which is short for pickle of course but with job

20
00:01:26,190 --> 00:01:28,010
Lib we do dot job lib.

21
00:01:28,660 --> 00:01:33,190
So now if we hit shift and enter there it's gonna give us little output here.

22
00:01:33,210 --> 00:01:39,630
Now if we go back to our working directory we'll see the same thing two seconds ago we have less random

23
00:01:39,630 --> 00:01:44,100
for us Model 1 job lib and again you can change this file name to whatever you want.

24
00:01:44,100 --> 00:01:49,020
See now we've got about half a dozen models just sitting in our working directory at this point we might

25
00:01:49,020 --> 00:01:53,710
move them into a model's file so we come back here.

26
00:01:53,710 --> 00:01:58,090
Now if we wanted to load in our job lib model you can probably guess which function we use to do that

27
00:01:59,720 --> 00:02:01,030
if not we're going to go through it.

28
00:02:01,080 --> 00:02:07,040
So we'll go import a saved job labor model and we're just going to import the exact same model that

29
00:02:07,040 --> 00:02:07,460
we save.

30
00:02:07,460 --> 00:02:13,190
So loaded job lib model equals load file name.

31
00:02:13,190 --> 00:02:19,020
We just have to pass it the exact same file name as what we saved it as G.S. random.

32
00:02:19,400 --> 00:02:22,450
We should be had to hit tab order complete dot job lib.

33
00:02:22,550 --> 00:02:26,170
Wonderful so that's going to load it in.

34
00:02:26,280 --> 00:02:27,510
Oh what have we got here.

35
00:02:27,520 --> 00:02:33,370
Oh we missed inverted comma full name making typos.

36
00:02:33,370 --> 00:02:39,820
Left right and center even using tab order complaint beautiful and again once the imported we can make

37
00:02:39,820 --> 00:02:41,590
predictions with our loaded model.

38
00:02:41,900 --> 00:02:56,590
So let's see it again make and evaluate job lib predictions job lib y reds equals loaded job leave model

39
00:02:56,650 --> 00:03:03,250
don't predict x test wonderful and then once again we'll use our evaluate Fred see how this has come

40
00:03:03,250 --> 00:03:08,710
in handy we defined this function a few videos ago but now we've used it at least half a dozen times

41
00:03:08,980 --> 00:03:14,860
and so preventing us from having to type out a bunch of different code the same code over and over again.

42
00:03:14,920 --> 00:03:18,250
So job lib y spreads shift and enter.

43
00:03:18,280 --> 00:03:26,920
We should get loaded job the model is not defined we're making this loaded job model might do change

44
00:03:26,920 --> 00:03:29,740
that to job lib and then this should work.

45
00:03:29,740 --> 00:03:30,610
Wonderful.

46
00:03:30,640 --> 00:03:33,190
So if we compare this this is with job lib.

47
00:03:33,190 --> 00:03:38,160
These are with pickle seventy eight point six nine tick points and four tick.

48
00:03:38,200 --> 00:03:40,090
Point eight to take point seven eight.

49
00:03:40,090 --> 00:03:40,790
Tick.

50
00:03:40,800 --> 00:03:44,540
So again the exact same thing with job lib as pickle.

51
00:03:45,100 --> 00:03:47,430
And now which one should you use.

52
00:03:47,800 --> 00:03:51,480
Well according to socket loans documentation let's have a look.

53
00:03:51,550 --> 00:03:56,950
So I kid loan exporting a model exporting a model.

54
00:03:56,950 --> 00:03:57,850
Model persistence.

55
00:03:57,850 --> 00:04:00,280
That's the word I was looking for.

56
00:04:00,300 --> 00:04:04,110
So again if you're thinking of something that you're trying to do with a library right.

57
00:04:04,110 --> 00:04:05,940
Just search the name a library.

58
00:04:05,940 --> 00:04:07,340
This is exactly what I do.

59
00:04:07,340 --> 00:04:09,620
I typed in cyclone saving a model.

60
00:04:09,690 --> 00:04:10,590
That's what I wanted to do.

61
00:04:10,650 --> 00:04:15,900
I'll come up to the documentation after training a psychic line model it is desirable to have a way

62
00:04:15,900 --> 00:04:19,050
to persist the model for future use without having to retrain.

63
00:04:19,170 --> 00:04:21,210
That's a main benefit of saving a model right.

64
00:04:22,110 --> 00:04:23,770
So here's an example.

65
00:04:23,770 --> 00:04:26,760
So you can use python's built in persistence model pickle.

66
00:04:26,760 --> 00:04:28,120
That's what we've seen.

67
00:04:28,380 --> 00:04:32,450
And now you could also use in the specific case of cyclone.

68
00:04:32,610 --> 00:04:38,520
It may be better to use job libs replacement of pickle dump and load which is more efficient on objects

69
00:04:38,520 --> 00:04:40,710
that carry large numbers higher res.

70
00:04:40,740 --> 00:04:46,530
Okay internally as is often the case for fitted psychic loan estimate is.

71
00:04:46,620 --> 00:04:48,780
And that's what we're saving in our case right.

72
00:04:48,810 --> 00:04:59,290
We're saving a fitted socket loan estimated a.k.a. G.S. CSF which is our fitted grid search random forest

73
00:04:59,290 --> 00:05:00,010
classifier

74
00:05:03,850 --> 00:05:07,160
but can only pixel to the disk and not to a string.

75
00:05:07,180 --> 00:05:13,150
So what this is saying is essentially sometimes if your model is large it might be more efficient because

76
00:05:13,330 --> 00:05:18,280
a model is made up of large number higher res a.k.a. patterns in numbers.

77
00:05:18,310 --> 00:05:21,280
That's what these num pi arrays are internally full of.

78
00:05:21,310 --> 00:05:25,600
That's what our classifier is probably full of just different num pi arrays with patterns between them.

79
00:05:25,600 --> 00:05:27,620
That's what it's learned from the data.

80
00:05:27,670 --> 00:05:32,950
So what this is saying is that if your model is large in our case our models are relatively small because

81
00:05:32,950 --> 00:05:36,010
we're only working with 300 or so samples.

82
00:05:36,400 --> 00:05:41,790
But if it is large it may be more efficient to use job leave over pickle.

83
00:05:42,310 --> 00:05:43,870
That's the main thing to worry about there.

84
00:05:44,850 --> 00:05:49,120
Alrighty so that's the crux of saving and loading a trained model.

85
00:05:49,120 --> 00:05:52,720
Again we've got them stored in in our directory here.

86
00:05:52,780 --> 00:05:57,220
If we needed to share this model that's something we could pass on to our production application or

87
00:05:57,220 --> 00:06:00,930
to our colleagues so they could replicate our work or something like that.

88
00:06:01,000 --> 00:06:04,520
Now we're really really on the homestretch.

89
00:06:04,540 --> 00:06:07,450
Next we're going to let's check back at our list.

90
00:06:07,450 --> 00:06:11,330
We're going to check out how we can put it all together.

91
00:06:11,350 --> 00:06:14,650
I won't spoil anything before the next video so I have a quick break.

92
00:06:14,650 --> 00:06:19,240
Try to say one of the models that we've trained using job liberal pickle and I'll see you in the next

93
00:06:19,240 --> 00:06:19,600
video.