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
|
|---|---|---|---|---|---|---|---|---|
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t291.92
|
up here, 128.
| 291.92 | 298.4 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t292.92
|
I'm also going to say how many, so how many results do we want to return?
| 292.92 | 301.4 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t298.4
|
I'm going to say 10.
| 298.4 | 303.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t301.4
|
Okay.
| 301.4 | 306.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t303.12
|
We also need to import FI's before we do anything.
| 303.12 | 308.74 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t306.12
|
And then we can initialize our index.
| 306.12 | 310.6 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t308.74
|
So I said we have two.
| 308.74 | 316.8 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t310.6
|
So we have FI's index flat 02 or IP.
| 310.6 | 319.8 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t316.8
|
I'm going to use IP because it's very slightly faster.
| 316.8 | 325.74 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t319.8
|
It seems from me testing it, it's very slightly faster, but there's hardly any difference
| 319.8 | 326.74 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t325.74
|
in reality.
| 325.74 | 329.88 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t326.74
|
So initializes our index and then we want to add our data to it.
| 326.74 | 334.16 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t329.88
|
So we add WB and then we perform a search.
| 329.88 | 344.48 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t334.16
|
So let me create a new cell and let me just run this quickly.
| 334.16 | 351.28 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t344.48
|
Okay.
| 344.48 | 355.4 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t351.28000000000003
|
And what I'm going to do is just time it so you can see how long this takes as well.
| 351.28 | 363.28 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t355.40000000000003
|
So I'm going to do time and we're going to do index or sorry, DI equals index search.
| 355.4 | 369.92 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t363.28
|
And in here we have our query vector and how many samples we'd like to return.
| 363.28 | 370.92 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t369.91999999999996
|
So I'm going to go with K.
| 369.92 | 377.36 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t370.91999999999996
|
Okay.
| 370.92 | 383 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t377.35999999999996
|
So that was reasonably quick and that's because we don't have a huge data set and we're just
| 377.36 | 384.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t383.0
|
searching for one query.
| 383 | 386.76 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t384.11999999999995
|
So it's not really too much of a problem there.
| 384.12 | 392.96 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t386.76
|
But what I do want to show you is, so if we print out I that returns all of the IDs or
| 386.76 | 398.16 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t392.96
|
the indexes of the 10 most similar vectors.
| 392.96 | 404.8 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t398.15999999999997
|
Now I'm going to use that as a baseline for each of our other indexes.
| 398.16 | 410.44 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t404.79999999999995
|
So this is, like I said, 100% quality and we can use this accuracy to test out other
| 404.8 | 411.86 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t410.44
|
indexes as well.
| 410.44 | 417.16 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t411.85999999999996
|
So what I'm going to do is take that and convert it into a list.
| 411.86 | 421 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t417.15999999999997
|
And if we just have a look at what we get, we see that we get a list like that.
| 417.16 | 427.84 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t421.0
|
And we're just going to use that, like I said, to see how our other indexes are performing.
| 421 | 429.6 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t427.84
|
So we'll move on to the other indexes.
| 427.84 | 434.24 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t429.6
|
And like I said before, we want to try and go from this, which is the flat indexes, which
| 429.6 | 438.04 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t434.24
|
is 100% search quality to something that's more 50-50.
| 434.24 | 439.74 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t438.04
|
But it depends on our use case as well.
| 438.04 | 443.2 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t439.74
|
Sometimes we might want more speed, sometimes higher quality.
| 439.74 | 448.84 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t443.2
|
So we will see a few of those through these indexes.
| 443.2 | 451.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t448.84
|
So we start with LSH.
| 448.84 | 453.56 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t451.11999999999995
|
So a very high level.
| 451.12 | 458.68 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t453.56
|
LSH works by grouping vectors in two different buckets.
| 453.56 | 464 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t458.67999999999995
|
Now what we can see on the screen now is a typical hashing function for like a Python
| 458.68 | 465.52 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t464.0
|
dictionary.
| 464 | 470.56 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t465.52
|
And what these hashing functions do is they try to minimize collisions.
| 465.52 | 478.16 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t470.55999999999995
|
So collision is where we would have the case of two items, maybe say these two, being hashed
| 470.56 | 479.64 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t478.16
|
into the same bucket.
| 478.16 | 484.72 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t479.64000000000004
|
And with a dictionary, you don't want that because you want every bucket to be an independent
| 479.64 | 485.72 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t484.72
|
value.
| 484.72 | 492.48 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t485.72
|
Otherwise, it increases the complexity of extracting your values from a single bucket
| 485.72 | 493.64 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t492.48
|
if they've collided.
| 492.48 | 498.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t493.64000000000004
|
Now LSH is slightly different because we actually do want to group things.
| 493.64 | 500.68 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t498.12
|
So we can see it as a dictionary.
| 498.12 | 506 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t500.68
|
But rather than where before we were avoiding those collisions, you can see here we're putting
| 500.68 | 509.36 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t506.0
|
them into completely different buckets every time.
| 506 | 512.44 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t509.36
|
Rather than doing that, we're trying to maximize collisions.
| 509.36 | 518.08 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t512.44
|
So you can see here that we've pushed all three of these keys into this single bucket
| 512.44 | 519.76 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t518.08
|
here.
| 518.08 | 523.06 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t519.76
|
And we've also pushed all of these keys into this single bucket.
| 519.76 | 525.32 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t523.06
|
So we get groupings of our values.
| 523.06 | 532.52 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t525.32
|
Now when it comes to performing our search, we process our query through the same hashing
| 525.32 | 536.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t532.52
|
function and that will push it to one of our buckets.
| 532.52 | 542.52 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t536.12
|
Now in the case of maybe appearing in this bucket here, we use Hamming Distance to find
| 536.12 | 552.48 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t542.52
|
the nearest bucket and then we can search or we restrict our scope to these values.
| 542.52 | 557.84 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t552.48
|
So we just restricted our scope there, which means that we do not need to search through
| 552.48 | 558.84 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t557.84
|
everything.
| 557.84 | 563.56 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t558.84
|
We are avoiding searching through those values down there.
| 558.84 | 566.66 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t563.5600000000001
|
Now let's have a look at how we implement that.
| 563.56 | 568 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t566.6600000000001
|
So it's pretty straightforward.
| 566.66 | 572.9 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t568.0
|
All we do is index, we do vise index LSH.
| 568 | 577.44 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t572.9
|
We have our dimensionality and then we also have this other variable which is called nbits.
| 572.9 | 582.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t577.44
|
So I will put that in a variable up here.
| 577.44 | 587.56 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t582.12
|
Do nbits and what I'm going to do is I'm going to make it d multiplied by 4.
| 582.12 | 593.44 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t587.56
|
So nbits we will have to scale with the dimensionality of our data which comes into another problem
| 587.56 | 597.96 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t593.4399999999999
|
which I'll mention later on which is the curse of dimensionality.
| 593.44 | 601.06 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t597.9599999999999
|
But I'll talk more about it in a moment.
| 597.96 | 607.72 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t601.06
|
So here we have nbits and then we add our data like we did before and then we can search
| 601.06 | 610.96 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t607.7199999999999
|
our data just like we did before.
| 607.72 | 623.24 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t610.96
|
So time and we do d pi equals index search and we are searching using our query, our
| 610.96 | 630.6 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t623.24
|
search query and we want to return 10 items.
| 623.24 | 637.12 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t630.6
|
So quicker speed, see here.
| 630.6 | 645.96 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t637.12
|
And what we can also do is compare the results to our 100% quality index or flat index and
| 637.12 | 654.72 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t645.96
|
we do that using numpy in 1D baseline i.
| 645.96 | 659.2 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t654.72
|
Okay so I'm just going to look at it visually here so we can see we have quite a lot of
| 654.72 | 660.2 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t659.2
|
matches here.
| 659.2 | 663.6 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t660.2
|
So plenty of trues, couple of falses, true, false, false, false.
| 660.2 | 672.04 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t663.6
|
So these are the top 10 that have been returned using our LSH algorithm and we're checking
| 663.6 | 679.08 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t672.0400000000001
|
if they exist in the baseline results that we got from our flat index earlier and we're
| 672.04 | 682.64 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t679.08
|
returning that most of them are present in that baseline.
| 679.08 | 686.68 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t682.64
|
So most of them do match so it's a reasonably good recall there.
| 682.64 | 688 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t686.6800000000001
|
So that's good and it was faster.
| 686.68 | 691.4 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t688.0
|
So we've got 17.6 milliseconds here.
| 688 | 693.08 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t691.4
|
How much did we get up here?
| 691.4 | 696.74 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t693.08
|
We got 157 milliseconds.
| 693.08 | 702.66 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t696.74
|
So slightly less accurate but what is that 10 times faster so it's pretty good.
| 696.74 | 704.96 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t702.6600000000001
|
And we can mess around with n bits.
| 702.66 | 710.36 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t704.96
|
We can increase it to increase the accuracy of our index or we decrease it to increase
| 704.96 | 711.72 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t710.36
|
the speed.
| 710.36 | 714.6 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t711.72
|
So again it's just trying to find that balance between them both.
| 711.72 | 721.58 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t714.6
|
Okay so this is a graph just showing you the recalls with different n bit values.
| 714.6 | 728.06 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t721.58
|
So as we sort of saw before we increase the n bits value for good recall but at the same
| 721.58 | 730 |
Choosing Indexes for Similarity Search (Faiss in Python)
|
2021-08-09 15:04:10 UTC
|
https://youtu.be/B7wmo_NImgM
|
B7wmo_NImgM
|
UCv83tO5cePwHMt1952IVVHw
|
B7wmo_NImgM-t728.0600000000001
|
time we have that curse of dimensionality.
| 728.06 | 738.32 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.