DocArray logo: The data structure for unstructured data

The data structure for unstructured data

Python 3.7 3.8 3.9 3.10
PyPI

DocArray is a library for nested, unstructured data such as text, image, audio, video, 3D mesh. It allows deep learning engineers to efficiently process, embed, search, recommend, store, transfer the data with Pythonic API.

? All data types: super-expressive data structure for representing complicated/mixed/nested text, image, video, audio, 3D mesh data.

? Pythonic experience: designed to be as easy as Python list. If you know how to Python, you know how to DocArray. Intuitive idioms and type annotation simplify the code you write.

?‍? Data science powerhouse: greatly accelerate data scientists work on embedding, matching, visualizing, evaluating via Torch/Tensorflow/ONNX/PaddlePaddle on CPU/GPU.

? Portable: ready-to-wire at anytime with efficient and compact serialization from/to Protobuf, bytes, JSON, CSV, dataframe.

Read more on why should you use DocArray and comparison to alternatives.

Install

Requires Python 3.7+ and numpy only:

pip install docarray

Additional features can be enabled by installing the full dependencies: pip install "docarray[full]".

Documentation

Get Started

DocArray consists of two simple concepts:

  • Document: a data structure for easily representing nested, unstructured data.
  • DocumentArray: a container for efficiently accessing, manipulating, and understanding multiple Documents.

A 10-liners text matching

We search for top-5 similar sentences of she smiled too much in “Pride and Prejudice”.

from docarray import Document, DocumentArray

d = Document(uri='https://www.gutenberg.org/files/1342/1342-0.txt').load_uri_to_text()
da = DocumentArray(Document(text=s.strip()) for s in d.text.split('\n') if s.strip())
da.apply(lambda d: d.embed_feature_hashing())

q = (Document(text='she smiled too much')
     .embed_feature_hashing()
     .match(da, metric='jaccard', use_scipy=True))

print(q.matches[:5, ('text', 'scores__jaccard__value')])

[['but she smiled too much.', 
  '_little_, she might have fancied too _much_.', 
  'She perfectly remembered everything that had passed in', 
  'tolerably detached tone. While she spoke, an involuntary glance', 
  'much as she chooses.”'], 
  [0.3333333333333333, 0.6666666666666666, 0.7, 0.7272727272727273, 0.75]]

Here the feature embedding is done by simple feature hashing and distance metric is Jaccard distance. You got better embedding? Of course you do! Looking forward to seeing your results.

A complete workflow of visual search

Let’s use DocArray and Totally Looks Like dataset to build simple meme image search. The dataset contains 6016 image-pairs stored in /left and /right. Images that shares the same filename are perceptually similar. For example,

left/00018.jpg right/00018.jpg left/00131.jpg right/00131.jpg
Visualizing top-9 matches using DocArray API Visualizing top-9 matches using DocArray API Visualizing top-9 matches using DocArray API Visualizing top-9 matches using DocArray API

Our problem is given an image from /left and find its most-similar image in /right (without looking at the filename of course).

Load images

First load images and preprocess them with standard computer vision techniques:

from docarray import DocumentArray, Document

left_da = DocumentArray.from_files('left/*.jpg')

To get a feeling of the data you will handle, plot them in one sprite image:

left_da.plot_image_sprites()

Load totally looks like dataset with docarray API

Apply preprocessing

Let’s do some standard computer vision preprocessing:

def preproc(d: Document):
    return (d.load_uri_to_image_blob()  # load
             .set_image_blob_normalization()  # normalize color 
             .set_image_blob_channel_axis(-1, 0))  # switch color axis for the pytorch model later

left_da.apply(preproc)

Did I mention apply work in parallel?

Embed images

Now convert images into embeddings using a pretrained ResNet50:

import torchvision
model = torchvision.models.resnet50(pretrained=True)  # load ResNet50
left_da.embed(model, device='cuda')  # embed via GPU to speedup

This step takes ~30 seconds on GPU. Beside PyTorch, you can also use Tensorflow, PaddlePaddle, ONNX models in .embed(...).

Visualize embeddings

You can visualize the embeddings via tSNE in an interactive embedding projector:

left_da.plot_embeddings()

Visualizing embedding via tSNE and embedding projector

Fun is fun, but recall our goal is to match left images against right images and so far we have only handled the left. Let’s repeat the same procedure for the right:

right_da = (DocumentArray.from_files('right/*.jpg')
                         .apply(preproc)
                         .embed(model, device='cuda'))

Match nearest neighbours

We can now match the left to the right and take the top-9 results.

left_da.match(right_da, limit=9)

Let’s inspect what’s inside left_da matches now:

for d in left_da:
    for m in d.matches:
        print(d.uri, m.uri, m.scores['cosine'].value)

left/02262.jpg right/03459.jpg 0.21102
left/02262.jpg right/02964.jpg 0.13871843
left/02262.jpg right/02103.jpg 0.18265384
left/02262.jpg right/04520.jpg 0.16477376
...

Or shorten the loop as one-liner:

print(left_da['@m', ('uri', 'scores__cosine__value')])

Better see it.

(DocumentArray(left_da[8].matches, copy=True)
    .apply(lambda d: d.set_image_blob_channel_axis(0, -1)
                      .set_image_blob_inv_normalization())
    .plot_image_sprites('result.png'))

Visualizing top-9 matches using DocArray API
Visualizing top-9 matches using DocArray API

What we did here is reverting the preprocessing steps (i.e. switching axis and normalizing) on the copied matches, so that one can visualize them using image sprites.

Quantitative evaluation

Serious as you are, visual inspection is surely not enough. Let’s calculate the recall@K. First we construct the groundtruth matches:

groundtruth = DocumentArray(
    Document(uri=d.uri, matches=[Document(uri=d.uri.replace('left', 'right'))]) for d in left_da)

Here we create a new DocumentArray with real matches by simply replacing the filename, e.g. left/00001.jpg to right/00001.jpg. That’s all we need: if the predicted match has the identical uri as the groundtruth match, then it is correct.

Now let’s check recall rate from 1 to 5 over the full dataset:

for k in range(1, 6):
    print(f'recall@{k}',
          left_da.evaluate(
            groundtruth,
            hash_fn=lambda d: d.uri,
            metric='recall_at_k',
            k=k,
            max_rel=1))

recall@1 0.02726063829787234
recall@2 0.03873005319148936
recall@3 0.04670877659574468
recall@4 0.052194148936170214
recall@5 0.0573470744680851

More metrics can be used such as precision_at_k, ndcg_at_k, hit_at_k.

If you think a pretrained ResNet50 is good enough, let me tell you with Finetuner one could do much better in just 10 extra lines of code. Here is how.

Save results

You can save a DocumentArray to binary, JSON, dict, dataframe, CSV or Protobuf message with/without compression. In its simplest form,

left_da.save('left_da.bin')

To reuse it, do left_da = DocumentArray.load('left_da.bin').

If you want to transfer a DoucmentArray from one machine to another or share it with your colleagues, you can do:

left_da.push(token='my_shared_da')

Now anyone who knows the token my_shared_da can pull and work on it.

left_da = DocumentArray.pull(token='my_shared_da')

Intrigued? That’s only scratching the surface of what DocArray is capable of. Read our docs to learn more.

Support

Join Us

DocArray is backed by Jina AI and licensed under Apache-2.0. We are actively hiring AI engineers, solution engineers to build the next neural search ecosystem in opensource.

GitHub

View Github