title
stringlengths 12
112
| published
stringlengths 19
23
| url
stringlengths 28
28
| video_id
stringlengths 11
11
| channel_id
stringclasses 5
values | id
stringlengths 16
31
| text
stringlengths 0
596
| start
float64 0
37.8k
| end
float64 2.18
37.8k
|
|---|---|---|---|---|---|---|---|---|
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t379.28
|
no worries.
| 379.28 | 388.4 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t380.16
|
And then we also need to specify our index and at the moment we don't have an Aurelius index and
| 380.16 | 394.96 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t388.40000000000003
|
that's fine because this will initialize it for us. So we'll just call it Aurelius.
| 388.4 | 406.64 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t400.0
|
Now if we go down here we can see what it actually did so it sent a put request to here.
| 400 | 415.52 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t406.64
|
localhost 9200 Aurelius. So that's how you create a new index. After that what we want to do is
| 406.64 | 425.76 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t416.24
|
first import our data. So we have the data here which I got from this website
| 416.24 | 434.48 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t425.76
|
and process with this script which you can find on GitHub. I'll keep a link in the description so you
| 425.76 | 440.8 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t434.48
|
can just go and copy that if you need to. Now I haven't really done much pre-process it's pretty
| 434.48 | 449.52 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t440.8
|
straightforward and all you need to do here is actually open that data. So we do that with open
| 440.8 | 458.88 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t449.52
|
and from here that data file is located two folders up in a data folder it's called meditations.txt.
| 449.52 | 462.56 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t461.52
|
I'm going to be reading that
| 461.52 | 470.64 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t466.71999999999997
|
and all we do is data equals f.read
| 466.72 | 478.72 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t470.64
|
and then if we just have a quick look at the first 100 characters there we see that we have
| 470.64 | 488.32 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t478.71999999999997
|
this newline character and that signifies a new paragraph from the text. So what we want to do here
| 478.72 | 496.96 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t490.47999999999996
|
is split the data and then we can see that we have a newline character.
| 490.48 | 502.24 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t496.96
|
So what we want to do is split the data by newline
| 496.96 | 511.52 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t504.64
|
and then if we check the length of that see that we have 508 separate paragraphs in there.
| 504.64 | 521.76 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t512.8
|
So what we now want to do is we want to modify this data so that it's in the correct format
| 512.8 | 531.2 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t521.76
|
for Haystack and Elasticsearch. So that format looks like this so it expects a list of dictionaries
| 521.76 | 540.4 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t531.2
|
where each dictionary looks like this from the text and inside here we would have our paragraph.
| 531.2 | 549.68 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t541.12
|
So each one of these items here and then there's another optional field called meta
| 541.12 | 553.76 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t549.68
|
and meta contains a dictionary and in here we can put whatever we want.
| 549.68 | 561.92 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t554.88
|
So for us I don't think at the moment there's really that much to put into here other than
| 554.88 | 569.44 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t561.92
|
where it came from so the book or maybe the source is probably a better word to use here
| 561.92 | 577.28 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t570.64
|
and all of these are coming from Meditations. Now later on we will probably add a few other
| 570.64 | 583.44 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t577.28
|
books as well and then the source will be different and when we return that item from
| 577.28 | 588.08 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t583.4399999999999
|
our retriever and our reader we'll at least be able to see which book came from him.
| 583.44 | 595.6 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t588.0799999999999
|
It would also be pretty cool to maybe include like a page number or something but at the moment with
| 588.08 | 601.2 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t595.6
|
this there are no page numbers included so we're not doing that at the moment.
| 595.6 | 609.36 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t601.2
|
So that's the format that we need and it's going to be a list of these.
| 601.2 | 614.08 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t610.5600000000001
|
So to do that we'll just do some list comprehension.
| 610.56 | 623.76 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t617.36
|
So we're going to write this and let's just copy this I think yeah it should be fine we'll copy this
| 617.36 | 634.64 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t623.76
|
and just indent that and in here we have our paragraph and sources Meditations for all of them
| 623.76 | 645.36 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t634.64
|
and then we just write for paragraph in and data okay so yeah that should work and if we just
| 634.64 | 655.44 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t645.36
|
and if we just check what we have here okay so that's that's what we want so we have text
| 645.36 | 659.92 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t655.44
|
we have the paragraph and then in here we have this meta with a source which is always Meditations
| 655.44 | 667.52 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t659.92
|
at the moment so that looks pretty good and we'll just double check the length again it should be
| 659.92 | 678.8 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t667.52
|
508 okay perfect now what we need to do is index all of these documents into our Elasticsearch
| 667.52 | 685.28 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t678.8
|
instance and to do that it's super easy all we do is call docstore because we're doing this through
| 678.8 | 698.16 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t685.28
|
Haystack now and we do write documents and we just pass in our data.json and that should work.
| 685.28 | 708.48 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t698.16
|
Okay cool so we can see here what it's done as it's sent a POST request to the Bulk API and sent
| 698.16 | 717.28 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t708.48
|
two of them I assume because it can only send so many documents at once so that's pretty cool and
| 708.48 | 726.48 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t717.28
|
now what I want to check is that we actually have 508 documents in our Elasticsearch instance
| 717.28 | 734.8 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t728.24
|
so to do that we're going to revert back to requests so we'll do requests.get
| 728.24 | 740.88 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t734.8
|
again go to our localhost
| 734.8 | 751.44 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t744.0
|
9200 and here we need to specify the index that we want to count the number of entries in
| 744 | 759.36 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t752.0
|
and then all we do is add count onto the end there and this will return a JSON object so we do this
| 752 | 766.32 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t759.36
|
so that we can see it and sure enough we have 508 items in that document store.
| 759.36 | 776.64 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t767.44
|
So if we head on back to our original plan so up here we had meditations we've now got that
| 767.44 | 789.04 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t778.8000000000001
|
and we've also set up the first part of our sack over here so Elastic now has meditations in there
| 778.8 | 797.36 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t789.04
|
so we can cross that off now the next step is setting up our retriever which we'll cover in the
| 789.04 | 819.6 |
How to Index Q&A Data With Haystack and Elasticsearch
|
2021-04-12 15:00:11 UTC
|
https://youtu.be/Vwq7Ucp9UCw
|
Vwq7Ucp9UCw
|
UCv83tO5cePwHMt1952IVVHw
|
Vwq7Ucp9UCw-t797.36
| 797.36 | 819.6 |
|
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t0.0
|
Hi, welcome to the video. Here we're going to have a look at how we can pre-train BERT.
| 0 | 14.64 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t6.88
|
So what I mean by pre-train is fine-tune BERT using the same approaches that are used to actually
| 6.88 | 22.8 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t14.64
|
pre-train BERT itself. So we would use these when we want to teach BERT to better understand the
| 14.64 | 32.24 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t22.8
|
style of language in our specific use cases. So we'll jump straight into it, but what we're going
| 22.8 | 39.76 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t32.24
|
to see is essentially two different methods applied together. So when we're pre-training,
| 32.24 | 47.2 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t39.760000000000005
|
we're using something called mass language modeling or MLM and also net sentence prediction or NSP.
| 39.76 | 53.36 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t47.2
|
Now in a few previous videos, I've covered all of these. So if you do want to go into a little more
| 47.2 | 58.72 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t53.36
|
depth, then I would definitely recommend having a look at those. But in this video, we're just
| 53.36 | 64.72 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t58.72
|
going to go straight into actually training a BERT model using both of those methods using
| 58.72 | 72.72 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t64.72
|
the pre-training class. So we need first to import everything that we need. So I'm going to
| 64.72 | 78 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t72.72
|
import requests because I'm going to use request download data we're using, which is from here. You
| 72.72 | 88.16 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t78.0
|
find a link in the description for that. And we also need to import our tokenizer and model classes
| 78 | 93.76 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t88.16
|
from transformers. So from transformers, we're going to import BERT tokenizer
| 88.16 | 104.8 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t93.76
|
and also BERT for pre-training. Now, like I said before, this BERT for pre-training class contains
| 93.76 | 114.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t104.80000000000001
|
both an MLM head and an NSP head. So once we have that, we also need to import torch as well. So let
| 104.8 | 123.92 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t114.4
|
me import torch. Once we have that, we can initialize our tokenizer and model. So we initialize
| 114.4 | 132.64 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t123.92
|
our tokenizer like this. So BERT tokenizer and from pre-trained. And we're going to be using the
| 123.92 | 145.44 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t132.64
|
BERT base uncased model. Obviously, you can use whichever BERT model you'd like. And for our model,
| 132.64 | 152.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t145.44
|
we have the BERT for pre-training class. So that's our tokenizer model. Now let's get our data.
| 145.44 | 158 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t153.27999999999997
|
Don't need to worry about that warning. It's just telling us that we need to train it, basically,
| 153.28 | 167.84 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t158.0
|
if we want to use it for inference predictions. So we get our data. We're going to pull it from
| 158 | 178.16 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t167.84
|
here. So let me copy that. And it's just requests.get. And paste that in there. And we should
| 167.84 | 184.56 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t178.16
|
see a 200 code. That's good. And so we just extracted data using the text attribute.
| 178.16 | 190.72 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t184.56
|
So text equals that. We also need to split it because it's a set of paragraphs that are split
| 184.56 | 197.68 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t190.72
|
by a new line character. And we can see those in here. Now we need to power data both for NSP and
| 190.72 | 204.96 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t197.68
|
MLM. So we'll go with NSP first. And to do that, we need to create a set of random sentences. So
| 197.68 | 213.92 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t204.96
|
sentence A and B. And then we need to create a set of random sentences. So we need to create a set
| 204.96 | 222 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t213.92
|
of random sentences. So sentence A and B, where the sentence B is not related to sentence A. We
| 213.92 | 228.8 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t222.0
|
need roughly 50% of those. And then the other 50% we want it to be sentence A is followed by sentence
| 222 | 237.36 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t228.79999999999998
|
B. So they are more coherent. So we're basically teaching BERT to distinguish between coherence
| 228.8 | 248 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t237.36
|
between sentences. So like long term dependencies. And we just want to be aware that within our text,
| 237.36 | 255.44 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t248.0
|
so we have this one paragraph that has multiple sentences. So we split by this. We have those. So
| 248 | 261.6 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t255.44000000000003
|
we need to create essentially a list of all of the different sentences that we have that we can just
| 255.44 | 267.92 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t261.6
|
pull from when we're creating our training data for NSP. Now to do that, we're going to
| 261.6 | 273.52 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t268.88
|
use this comprehension here. And what we do is write sentence. So for each sentence,
| 268.88 | 278.88 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t274.40000000000003
|
for each paragraph in the text. So this variable.
| 274.4 | 289.04 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t282.64000000000004
|
For sentence in para.split. So this is where we're getting our sentence variable from.
| 282.64 | 296.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t289.04
|
And we just want to be aware of if we have a look at this one, we see we get this empty sentence,
| 289.04 | 300.96 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t296.40000000000003
|
we get that for all of our paragraphs. So we want to not include those. So we say if
| 296.4 | 310.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t301.68
|
sentence is not equal to that empty sentence. And we're also going to need to get the length
| 301.68 | 317.12 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t310.40000000000003
|
of that bag for later as well. And now what we do is create our NSP training data.
| 310.4 | 324.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t317.12
|
So we want that 50-50 split. So we're going to use the random library to create that 50-50 randomness.
| 317.12 | 333.04 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t326.24
|
We want to initialize a list of sentence a's, a list of sentence b's,
| 326.24 | 342.32 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t335.04
|
and also a list of labels. And then what we do is we're going to loop through each paragraph in
| 335.04 | 350.48 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t342.32
|
our text. So for paragraph in text. We want to extract each sentence from the paragraph. So
| 342.32 | 355.84 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t350.48
|
we're going to use it similar to what we've done here. So write sentences. This is going to be a
| 350.48 | 365.76 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t355.84
|
list of all the sentences within each paragraph. So sentence for sentence in para.split
| 355.84 | 374.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t365.76
|
by a period character. And we also want to make sure we're not including those empty ones. So if
| 365.76 | 384.64 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.