1
00:00:00,040 --> 00:00:05,840
Okay, so back to calling our favorite evaluate script with the ensemble function right here.
好的，回到这里使用 ensemble 函数调用我们最喜欢的评估脚本。

2
00:00:06,160 --> 00:00:09,600
And with our test data set I will kick it off.
我将通过我们的测试数据集开始它。

3
00:00:09,880 --> 00:00:11,720
And this time it will take longer.
而这一次需要的时间会更长。

4
00:00:11,720 --> 00:00:16,680
Particularly the first one is going to take the whole 30s while while modal kicks into gear.
特别是第一个将花费整个 30 秒的时间，同时模态启动。

5
00:00:16,960 --> 00:00:20,520
But once it's got through the first the first one, it will then speed up.
但一旦通过了第一个，它就会加速。

6
00:00:20,840 --> 00:00:25,960
So I'm going to let this do its thing, and hopefully you will do the same because you're going to want
所以我会让它发挥作用，希望你也会这样做，因为你会想要

7
00:00:25,960 --> 00:00:27,040
to try this yourself.
自己尝试一下。

8
00:00:27,360 --> 00:00:33,120
But I will say if for some of you that you might get technical problems running the deep neural network
但我想说的是，对于你们中的一些人来说，运行深度神经网络可能会遇到技术问题

9
00:00:33,120 --> 00:00:39,440
because it's quite, quite a hefty piece of machinery, if you want, you don't need to run this and
因为它是一个非常非常庞大的机器，如果你愿意的话，你不需要运行它

10
00:00:39,480 --> 00:00:40,240
are very easy.
都很容易。

11
00:00:40,280 --> 00:00:43,280
I mentioned before that there are some easy shortcuts.
我之前提到过有一些简单的捷径。

12
00:00:43,280 --> 00:00:48,320
If you have any problems running the ensemble model, you can just replace the ensemble model by something
如果运行集成模型时遇到任何问题，您可以将集成模型替换为其他模型

13
00:00:48,320 --> 00:00:51,640
that just makes simply the call to the specialist model only.
这只是简单地调用专家模型。

14
00:00:51,640 --> 00:00:53,520
There's no need to do this if you don't want.
如果您不愿意，则无​​需这样做。

15
00:00:53,560 --> 00:00:57,240
Anyway, it's off and running and I'm going to let it go through.
不管怎样，它已经开始运行了，我会让它继续下去。

16
00:00:57,280 --> 00:01:02,820
You can see once it's once modal has woken up, it's then pretty quick to call modal repeatedly.
您可以看到，一旦模态被唤醒，就可以很快地重复调用模态。

17
00:01:03,060 --> 00:01:07,580
So just in case it's not clear to you what we're doing, every time we come up with a price, we are
因此，以防万一您不清楚我们在做什么，每次我们提出价格时，我们都会

18
00:01:07,580 --> 00:01:10,060
calling the specialist model in modal.
在模态中调用专家模型。

19
00:01:10,060 --> 00:01:16,460
In the cloud, we're calling the we're calling our rag frontier model looking up in chroma, making
在云中，我们调用我们的抹布边界模型在色度中查找，使得

20
00:01:16,460 --> 00:01:18,340
that prompt to GPT five one.
提示 GPT 五一。

21
00:01:18,660 --> 00:01:23,660
And then we are also calling a deep neural network and having it estimate as well.
然后我们也调用深度神经网络并对其进行估计。

22
00:01:23,660 --> 00:01:25,060
And we're more than halfway through.
我们已经完成一半多了。

23
00:01:25,060 --> 00:01:27,460
And I will see you in a second when it completes.
当它完成时我会看到你。

24
00:01:27,460 --> 00:01:32,780
And I should have said it's of course, of those three prices, it's taking 80% times the frontier,
我应该说，当然，在这三个价格中，它占据了边界的 80% 倍，

25
00:01:32,780 --> 00:01:37,460
one plus 10% plus 10% that's being combined to this number.
一加 10% 再加上 10% 就得到了这个数字。

26
00:01:37,660 --> 00:01:38,860
Let's reveal.
让我们来揭晓吧。

27
00:01:39,100 --> 00:01:40,220
Here you have it.
在这里你有它。

28
00:01:40,260 --> 00:01:41,780
We got a two handle.
我们有两个手柄。

29
00:01:41,820 --> 00:01:45,100
We only made a slight improvement on the frontier number.
我们只在边界数字上做了轻微的改进。

30
00:01:45,100 --> 00:01:48,620
But we're down to 29.9.
但我们已经降至 29.9。

31
00:01:48,780 --> 00:01:51,500
And here the R-squared is now at 87%.
这里 R 平方现在为 87%。

32
00:01:51,660 --> 00:01:53,940
This visual looks beautiful.
这个视觉效果看起来很漂亮。

33
00:01:53,980 --> 00:01:55,500
Absolutely beautiful.
绝对美丽。

34
00:01:55,500 --> 00:01:58,730
Look how close everything is this now?
看看现在一切有多接近？

35
00:01:58,730 --> 00:01:59,370
We really are.
我们确实是。

36
00:01:59,410 --> 00:02:00,450
We're done with pricing.
我们已经完成了定价。

37
00:02:00,490 --> 00:02:04,330
If you're fed up with these charts, this is the final, final one.
如果您厌倦了这些图表，这是最后一张。

38
00:02:04,330 --> 00:02:07,410
I was so determined to get a two handle and I got there in the end.
我下定决心要获得两个手柄，最终我做到了。

39
00:02:07,610 --> 00:02:08,970
Write this down please.
请写下来。

40
00:02:08,970 --> 00:02:12,810
Your final 129 .90.
您的最终成绩为 129 .90。

41
00:02:13,090 --> 00:02:15,170
That's that is success.
这就是成功。

42
00:02:15,330 --> 00:02:20,330
Now I also I'm keenly aware of what you probably already thought, which is that these do all have a
现在我也敏锐地意识到你可能已经想到的，那就是这些确实都有一个

43
00:02:20,330 --> 00:02:21,290
plus and minus.
加号和减号。

44
00:02:21,290 --> 00:02:26,530
So overly fixating on the fact that it's 29 is perhaps silly because there is, of course, an error
因此，过分关注 29 这个事实可能是愚蠢的，因为这当然是一个错误

45
00:02:26,530 --> 00:02:27,170
bar around it.
周围有酒吧。

46
00:02:27,170 --> 00:02:28,090
But it feels good.
但感觉很好。

47
00:02:28,130 --> 00:02:30,770
You know, it feels nice to have got 29.9.
你知道，得到 29.9 的感觉很好。

48
00:02:30,810 --> 00:02:34,850
But the thing is, this is just based on a quick, broad setting of these weights.
但问题是，这只是基于对这些权重的快速、广泛的设置。

49
00:02:34,850 --> 00:02:39,290
If you have the time to build a proper ensemble model, you can do a lot better.
如果您有时间构建适当的集成模型，您可以做得更好。

50
00:02:39,290 --> 00:02:42,050
And by the way, you can throw in more models in here.
顺便说一句，您可以在此处添加更多模型。

51
00:02:42,050 --> 00:02:43,810
You could throw in other frontier models.
你可以加入其他前沿模型。

52
00:02:43,810 --> 00:02:46,770
You could put in the whole gamut of the different models that we came up with.
您可以将我们提出的所有不同模型放入其中。

53
00:02:46,810 --> 00:02:49,930
Throw back in XGBoost, throw back in the random forest.
回到 XGBoost，回到随机森林。

54
00:02:49,930 --> 00:02:53,330
I think I did the random forest last year instead of the neural network.
我想我去年做了随机森林而不是神经网络。

55
00:02:53,330 --> 00:02:54,170
So put in.
所以放进去。

56
00:02:54,410 --> 00:02:56,010
The more that you put in, the better.
你投入的越多越好。

57
00:02:56,050 --> 00:03:00,950
The thing about using linear regression for your ensemble is that linear regression is smart, and if
在集成中使用线性回归的好处是线性回归很聪明，并且如果

58
00:03:00,950 --> 00:03:04,470
something doesn't add any signal, it's not going to it's not going to include it.
有些东西不会添加任何信号，它不会包含它。

59
00:03:04,510 --> 00:03:05,830
It just won't get counted.
只是不会被计算在内。

60
00:03:05,830 --> 00:03:08,270
So you can put in as many models as you want.
因此您可以添加任意数量的模型。

61
00:03:08,350 --> 00:03:13,750
And linear regression will figure out how can I take some combination of these different models to give
线性回归将找出我如何将这些不同模型的组合给出

62
00:03:13,750 --> 00:03:15,670
you the best final outcome?
你最后最好的结局是什么？

63
00:03:15,670 --> 00:03:21,190
So you should be able to beat me with this 29 and that of course, is the challenge for you.
所以你应该能够用这个 29 击败我，这当然是对你的挑战。

64
00:03:21,310 --> 00:03:22,550
Use all of the techniques.
使用所有的技术。

65
00:03:22,550 --> 00:03:25,750
Use everything we've done here to get even better.
利用我们在这里所做的一切来变得更好。

66
00:03:25,950 --> 00:03:28,150
Get it, get it down into the 20s.
得到它，把它降到20多岁。

67
00:03:28,190 --> 00:03:29,190
Let's see what you could do.
让我们看看你能做什么。

68
00:03:29,470 --> 00:03:33,230
Last year, the best score that I got was 46 or something like that.
去年，我得到的最好成绩是46分左右。

69
00:03:33,230 --> 00:03:34,470
It might have been 47.
可能是47。

70
00:03:34,590 --> 00:03:39,110
Uh, so this I mean, honestly, my, my, I'm blown away by this result.
呃，所以我的意思是，老实说，我的，我的，我对这个结果感到震惊。

71
00:03:39,110 --> 00:03:40,230
I'm so happy.
我太高兴了。

72
00:03:40,270 --> 00:03:42,070
I wanted to get to a three handle.
我想要一个三手柄。

73
00:03:42,070 --> 00:03:43,510
That was that was my dream.
那就是我的梦想。

74
00:03:43,510 --> 00:03:48,030
And I got to a two handle 29.9 uh, really, really cool.
我得到了两个手柄 29.9 呃，真的，真的很酷。

75
00:03:48,150 --> 00:03:55,330
Uh, so, uh, yeah, please take a moment to to to celebrate this success, wallow in the glory of
呃，所以，呃，是的，请花一点时间来庆祝这一成功，沉浸在

76
00:03:55,330 --> 00:03:57,530
our $29.9 error.
我们的错误为 29.9 美元。

77
00:03:57,650 --> 00:04:01,010
And of course, it wouldn't be real if we weren't able to visualize that.
当然，如果我们无法想象这一点，它就不会是真实的。

78
00:04:01,010 --> 00:04:02,450
So I have to do this.
所以我必须这样做。

79
00:04:02,610 --> 00:04:03,090
Here we go.
开始了。

80
00:04:03,130 --> 00:04:04,690
Put it put it into this chart.
把它放进这个图表里。

81
00:04:04,810 --> 00:04:08,490
And here is the final result, the prediction error from each model.
这是最终结果，即每个模型的预测误差。

82
00:04:08,770 --> 00:04:15,330
Uh, we've got, uh, the original, the original, uh, machine learning stuff, the, the disastrous
呃，我们有，呃，原始的，原始的，呃，机器学习的东西，灾难性的

83
00:04:15,370 --> 00:04:19,010
ad performance right here with 87.6.
广告效果为 87.6。

84
00:04:19,010 --> 00:04:19,530
Crazy.
疯狂的。

85
00:04:19,730 --> 00:04:23,370
Uh, this was the first, uh, neural network we did that was at 63.
呃，这是我们做的第一个神经网络，当时是 63。

86
00:04:23,410 --> 00:04:24,930
These were all the frontier models.
这些都是前沿型号。

87
00:04:24,930 --> 00:04:28,730
The best was GPT five one at 44.74.
最好的是 GPT 五一，为 44.74。

88
00:04:29,170 --> 00:04:31,890
Then that was the disappointing fine tuning results.
然后就是令人失望的微调结果。

89
00:04:32,010 --> 00:04:38,650
This was our better deep neural network at 46, uh, and 46.49.
这是我们更好的深度神经网络，为 46，呃，46.49。

90
00:04:38,690 --> 00:04:45,970
Then that was the baseline, the worst of everything, all the way down to the fine tuned 39.85, which
然后这是基线，所有事情中最差的，一直到微调的 39.85，这

91
00:04:45,970 --> 00:04:47,130
was our winner.
是我们的赢家。

92
00:04:47,170 --> 00:04:48,650
Up until today.
直到今天。

93
00:04:48,930 --> 00:04:57,110
Now in comes GPT five one rag beating everything at 30.19, and then just coming in a hair lower is
现在，GPT 5 1 rag 以 30.19 击败了所有对手，然后只低了一点点

94
00:04:57,110 --> 00:04:59,910
our ensemble at 29.9.
我们的乐团在29.9。

95
00:04:59,950 --> 00:05:04,990
Combining this plus this one plus this one.
结合这个加上这个加上这个。

96
00:05:05,190 --> 00:05:08,230
And as I say, there's definitely more mileage here.
正如我所说，这里肯定有更多的里程。

97
00:05:08,230 --> 00:05:11,550
There's more juice and I, I can't wait to hear how you do it.
还有更多的果汁，我迫不及待地想听听你是怎么做的。

98
00:05:11,590 --> 00:05:17,830
And you must send me screenshots, post screenshots of that chart showing what what performance you're
你必须向我发送屏幕截图，发布该图表的屏幕截图，显示你的表现

99
00:05:17,830 --> 00:05:18,750
able to achieve.
能够实现。

100
00:05:18,990 --> 00:05:23,310
But for now, I'm going to be very, very happy with the 29.9.
但现在，我对 29.9 感到非常非常满意。

101
00:05:23,310 --> 00:05:23,910
Okay.
好的。

102
00:05:24,030 --> 00:05:27,990
Uh, but then just to make me really happy, we have to put this into an agent.
呃，但是为了让我真正高兴，我们必须将其放入代理中。

103
00:05:28,190 --> 00:05:35,230
Uh, and, uh, so we're going to have a first of all, we're going to try a frontier agent, which
呃，呃，所以我们首先要尝试一名边境特工，

104
00:05:35,230 --> 00:05:38,710
is one that that is basically going to call this whole rag flow.
这基本上就是所谓的整个抹布流。

105
00:05:38,990 --> 00:05:41,910
Um, so let's just check it out again.
嗯，那我们再检查一下吧。

106
00:05:41,910 --> 00:05:44,150
It's in the agents directory.
它位于代理目录中。

107
00:05:44,470 --> 00:05:46,670
There is frontier Agent.
有边境代理。

108
00:05:46,670 --> 00:05:47,470
Where are you?
你在哪里？

109
00:05:48,070 --> 00:05:49,870
Um, there it is.
嗯，就在那里。

110
00:05:50,070 --> 00:05:57,100
And the frontier agent is It's just exactly the same now as we as we look to to turn notebooks into,
前沿代理是，它与我们现在希望将笔记本变成的完全相同，

111
00:05:57,340 --> 00:05:58,700
uh, Python modules.
呃，Python 模块。

112
00:05:58,700 --> 00:06:04,780
And I always do recommend you start with a notebook, because it's just a great way to be in the experimental
我总是建议您从笔记本开始，因为这是进行实验的好方法

113
00:06:04,780 --> 00:06:06,260
mindset and iterating on things.
心态和对事物的迭代。

114
00:06:06,260 --> 00:06:09,340
But there comes a point when it's time to put things into Python modules.
但到了某个时刻，就需要将东西放入 Python 模块中了。

115
00:06:09,340 --> 00:06:10,580
And this is how I'm trying to show you.
这就是我试图向您展示的方式。

116
00:06:10,620 --> 00:06:12,580
Like production ready code.
就像生产就绪的代码一样。

117
00:06:12,620 --> 00:06:15,380
It means that it has things like, you know, docstrings.
这意味着它有诸如文档字符串之类的东西。

118
00:06:15,380 --> 00:06:17,060
So we've got, uh, good comments here.
所以我们在这里得到了，嗯，很好的评论。

119
00:06:17,060 --> 00:06:18,660
And I'm also using type hints.
我还使用类型提示。

120
00:06:18,660 --> 00:06:24,900
So I'm doing all the better practices that the data scientists in me has perhaps not, not been as careful
所以我正在做所有更好的实践，而我的数据科学家可能没有，没有那么小心

121
00:06:24,900 --> 00:06:25,260
with.
和。

122
00:06:25,260 --> 00:06:29,500
But now we're into more of the engineering side of being an AI engineer.
但现在我们更多地关注人工智能工程师的工程方面。

123
00:06:29,500 --> 00:06:31,380
We're looking to be a bit more careful with this.
我们希望对此更加小心。

124
00:06:31,700 --> 00:06:35,140
You can see that the frontier agent is just following exactly the same code, though.
不过，您可以看到边境代理只是遵循完全相同的代码。

125
00:06:35,220 --> 00:06:40,860
Now we've experimented on it and perfected it, we can turn it into a nice module, a nice class frontier
现在我们已经对它进行了实验并完善了它，我们可以将它变成一个很好的模块，一个很好的类前沿

126
00:06:40,900 --> 00:06:41,420
agent.
代理人。

127
00:06:41,620 --> 00:06:45,580
It has the same thing making the contacts, getting the messages for finding similars.
它具有相同的功能，即建立联系、获取寻找相似信息的消息。

128
00:06:45,700 --> 00:06:48,700
And then this is when it actually calls the price.
然后这就是它实际定价的时候。

129
00:06:48,900 --> 00:06:53,000
And so with that background, we can now simply use this.
有了这个背景，我们现在就可以简单地使用它了。

130
00:06:53,040 --> 00:06:57,560
We can call the Quadcast HyperX condenser Mic and find out how much it costs.
我们可以致电 Quadcast HyperX 电容麦克风并了解其价格。

131
00:06:57,760 --> 00:07:00,520
And it predicts that it costs $140.
预计售价为 140 美元。

132
00:07:00,560 --> 00:07:03,560
And I think I mentioned before that it actually costs me $130.
我想我之前提到过它实际上花了我 130 美元。

133
00:07:03,680 --> 00:07:06,560
Uh, so this is a really very, very good pricing.
呃，所以这是一个非常非常好的定价。

134
00:07:06,560 --> 00:07:09,600
And our specialist model got 90, which wasn't as good.
我们的专业模型得到了 90 分，这不是那么好。

135
00:07:09,760 --> 00:07:11,280
Uh, so, uh, yeah.
呃，所以，呃，是的。

136
00:07:11,320 --> 00:07:12,960
There there you have it.
就在那里，你就有了。

137
00:07:13,240 --> 00:07:16,760
Now, uh, one, one thing that, uh, that.
现在，呃，一件事，呃，那个。

138
00:07:16,800 --> 00:07:17,120
Oh, yeah.
哦，是的。

139
00:07:17,120 --> 00:07:17,960
Let's try this as well.
我们也试试这个。

140
00:07:18,160 --> 00:07:18,880
Let's try it.
我们来试试吧。

141
00:07:18,920 --> 00:07:21,280
Pricing a Shure Mv7 plus.
舒尔 Mv7 plus 的定价。

142
00:07:21,560 --> 00:07:24,000
Uh, now you might ask what that is.
呃，现在你可能会问那是什么。

143
00:07:24,000 --> 00:07:26,960
That is the microphone that I'm speaking to you on right now.
这就是我现在正在用的麦克风。

144
00:07:27,160 --> 00:07:31,040
Uh, that that is my current microphone, and that costs me 2.99.
呃，那是我现在的麦克风，花了我 2.99 美元。

145
00:07:31,200 --> 00:07:33,320
Uh, so, uh, yeah, it's very close.
呃，所以，呃，是的，非常接近。

146
00:07:33,320 --> 00:07:37,000
Again, it's, uh, obviously really knows this stuff.
再说一次，呃，它显然真的知道这些东西。

147
00:07:37,200 --> 00:07:45,160
Um, and we can also then, uh, have a neural network agent and this shows in purple.
嗯，然后我们也可以，呃，有一个神经网络代理，它显示为紫色。

148
00:07:45,400 --> 00:07:47,240
And so let's have a look at the neural network agent.
让我们看一下神经网络代理。

149
00:07:47,240 --> 00:07:47,800
Here it is.
这里是。

150
00:07:47,800 --> 00:07:49,660
It's very short and simple.
它非常简短。

151
00:07:49,660 --> 00:07:52,500
It just calls inference on our neural network.
它只是调用我们的神经网络的推理。

152
00:07:52,700 --> 00:07:55,940
Uh, and after after loading it in and setting it up.
呃，在加载并设置之后。

153
00:07:56,180 --> 00:08:00,100
So if we come back here, we can also do the same thing.
所以如果我们回到这里，我们也可以做同样的事情。

154
00:08:00,100 --> 00:08:01,180
And it doesn't do as well.
但效果并不好。

155
00:08:01,180 --> 00:08:02,860
It says 162.
上面写着162。

156
00:08:03,020 --> 00:08:06,060
Um, but it's still in the general ballpark.
嗯，但它仍然在一般范围内。

157
00:08:06,620 --> 00:08:14,260
And then finally we can start creating an ensemble agent to be using all three and combining them.
最后，我们可以开始创建一个集成代理来使用所有三个并将它们组合起来。

158
00:08:14,300 --> 00:08:17,980
So let's go ahead and have a look at the specialists, the ensemble agent right now.
现在让我们来看看专家、乐团经纪人。

159
00:08:18,020 --> 00:08:22,700
It's it's here in ensemble agent right there.
它就在乐团特工那里。

160
00:08:22,900 --> 00:08:24,940
So it's nice and short again.
所以它又好又短。

161
00:08:24,980 --> 00:08:28,660
But there is one thing about this that maybe you were expecting.
但有一件事可能是你所期待的。

162
00:08:28,700 --> 00:08:32,340
If you, if you spotted this, this thing that I, that I almost didn't mention.
如果你，如果你发现了这个，我几乎没有提到的这件事。

163
00:08:32,780 --> 00:08:34,140
So what does this do?
那么这有什么作用呢？

164
00:08:34,300 --> 00:08:40,620
It creates a specialist agent, a frontier agent, and a neural network agent in its init method.
它在其 init 方法中创建了一个专家代理、一个前沿代理和一个神经网络代理。

165
00:08:40,660 --> 00:08:47,400
It creates those three agents that it's delegating to, but it does also create the preprocessor.
它创建了它所委托的三个代理，但它也创建了预处理器。

166
00:08:47,400 --> 00:08:48,640
Remember, we still have to.
请记住，我们仍然必须这样做。

167
00:08:48,680 --> 00:08:53,120
If we're taking in some text from the internet, we still have to pre-process that to rewrite that in
如果我们从互联网上获取一些文本，我们仍然需要对其进行预处理以将其重写为

168
00:08:53,120 --> 00:08:56,440
the same format that is going to be most suitable for our agents.
最适合我们代理商的相同格式。

169
00:08:56,720 --> 00:09:04,480
So it's then going to take the, the, the, the, the text that comes in in this price function, take
因此，它将采用此价格函数中出现的文本，采用

170
00:09:04,480 --> 00:09:08,760
the text and first use the preprocessor to rewrite it.
文本并首先使用预处理器重写它。

171
00:09:09,200 --> 00:09:11,160
Then it will call the specialist.
然后它会呼叫专家。

172
00:09:11,200 --> 00:09:16,960
The frontier and the neural network are three prices those three agents to the right.
边界和神经网络是右边这三个代理的三个价格。

173
00:09:17,320 --> 00:09:22,880
And it combines them by combining by multiplying the frontier by 0.8, the specialist by 0.1, and the
它通过将前沿乘以 0.8、将专家乘以 0.1 以及将

174
00:09:22,880 --> 00:09:24,400
neural network by 0.1.
神经网络 0.1。

175
00:09:24,680 --> 00:09:29,000
And that is what it returns, and it logs that as well.
这就是它返回的内容，并记录下来。

176
00:09:29,600 --> 00:09:33,280
If you're going to come in and improve the ensemble, this is where to do it.
如果你想加入并改进合奏，这里就是你应该做的地方。

177
00:09:33,320 --> 00:09:37,360
Of course, this is where you could use real weights that we come up with by looking at the data.
当然，您可以使用我们通过查看数据得出的实际权重。

178
00:09:37,400 --> 00:09:42,240
Look at the val, the validation data set perhaps to come up with the best way to weight it.
看看val，验证数据集也许能想出最好的加权方法。

179
00:09:42,760 --> 00:09:47,790
Also, if you have technical problems getting the neural network to work, or getting one of either
另外，如果您在使神经网络正常工作或获取其中之一方面遇到技术问题

180
00:09:47,830 --> 00:09:49,670
the specialist or the frontier to work.
专家或前沿工作。

181
00:09:49,710 --> 00:09:51,430
This is a shortcut you can take.
这是您可以采取的捷径。

182
00:09:51,470 --> 00:09:56,310
You can come into this ensemble agent, and you can just comment out whichever ones give you trouble.
你可以进入这个集成代理，你可以直接注释掉那些给你带来麻烦的人。

183
00:09:56,310 --> 00:10:00,190
Just comment them out here, comment them out here and then down here.
只需在这里注释掉它们，在这里注释掉它们，然后在这里注释掉。

184
00:10:00,190 --> 00:10:03,830
Comment them out and just return the number that you actually want.
将它们注释掉并返回您真正想要的数字。

185
00:10:03,910 --> 00:10:11,310
Let me uncomment it up so that that would allow you to simplify the ensemble agent to do whatever it
让我取消注释，这样您就可以简化集成代理来执行任何操作

186
00:10:11,310 --> 00:10:11,950
is that you want.
就是你想要的。

187
00:10:11,950 --> 00:10:15,310
This is one place to go to, to really pare down what's happening.
这是一个可以去的地方，可以真正减少正在发生的事情。

188
00:10:15,310 --> 00:10:17,910
If you want to, to run a minimal version of this.
如果你愿意，可以运行它的最小版本。

189
00:10:18,070 --> 00:10:20,790
Um, but but we'll, we'll have the maximum version of this.
嗯，但是我们会，我们会拥有这个的最大版本。

190
00:10:20,830 --> 00:10:24,670
We'll, we'll, we'll be using this combined weighting.
我们将、我们将、我们将使用这种组合权重。

191
00:10:24,710 --> 00:10:30,870
And you'll notice this is an example of really an agent workflow that we're just using Python code to
您会注意到，这是一个真正的代理工作流程示例，我们只是使用 Python 代码来实现

192
00:10:30,910 --> 00:10:33,950
orchestrate the three calls to three different agents.
将三个呼叫协调到三个不同的代理。

193
00:10:33,950 --> 00:10:36,190
It's not like there's any autonomy at this point.
目前看来还没有任何自主权。

194
00:10:36,190 --> 00:10:41,910
It's just a fixed calling for three sub agents in order to combine those numbers and come up with the
这只是对三个子代理的固定调用，以便将这些数字组合起来并得出

195
00:10:41,910 --> 00:10:44,510
total estimated price of this item.
该商品的预估总价。