Smahat Time Series Encoding
Smahat allows to encode a sequence of integer values using a fixed (for all values) number of bits but minimal with regards to the data range. For example: for a series of boolean values only one bit is needed, for a series of integer percentages 7 bits are needed, etc.
Smahat is useful when:
- Time series is integer-valued. (It doesn't work with floats :))
- The range of the data is known in advance (if not streaming, this is not necessary).
- The data range is relatively small.
- The data does not have properties that would make other compression algorithms useful, or these other algorithms have an unacceptable cost for the use case.
Smahat can also be used as a baseline to calculate the true compression ratio of a compression algorithm on data of a certain nature.
Installation
To install the latest release:
$ pip install smahat
You can also build a local package and install it:
$ make build
$ pip install dist/*.whl
Usage
Import smahat module.
>>> import smahat
Data to encode.
>>> values = [12, 0, 17, 15, 78, 10]
Encoding
You can use encode_next to encode one value by one:
>>> encoder = smahat.Encoder(min_value=0, max_value=100, strategy='saturate')
>>> for v in values:
... encoder.encode_next(v)
>>> content = encoder.get_encoded()
>>> content
{'encoded': b'\x18\x00\x88\xf9\xc2\x80', 'shift': 0, 'n_bits_per_value': 7, 'n_padding_bits': 6}
Or you can use Encoder.encode_all to encode all values (range min and max will be inferred from values if not provided):
>>> content = smahat.Encoder.encode_all(values, min_value=0, max_value=100, strategy='saturate')
>>> content
{'encoded': b'\x18\x00\x88\xf9\xc2\x80', 'shift': 0, 'n_bits_per_value': 7, 'n_padding_bits': 6}
Decoding
To decode use Decoder.decode_all.
>>> smahat.Decoder.decode_all(content)
[12, 0, 17, 15, 78, 10]
Encoding result
The result of the encoding of a sequence of values using Smahat is a SmahatContent dictionary containing the encoded data, plus three fields : shift is used to bring the data range to start from zero (values are shifted and encoded in pure binary), n_bits_per_value indicates the number of bits used to encode each value, n_padding_bits (between 0 and 7) indicates the number of unused padding bits within the last byte.
class SmahatContent(TypedDict):
encoded: bytes
shift: int
n_bits_per_value: int
n_padding_bits: int
If you want to use this library for message exchanges, you can serialize the result of the encoding as you like (JSON, protobuf, etc.)
Contribute
$ git clone https://github.com/ghilesmeddour/smahat-time-series-encoding.git
$ cd smahat-time-series-compression
make format
make dead-code-check
make test
make type-check
make coverage
make build
TODOs
- Add unit tests.
- Improve doc.
712 Dec 23, 2022
638 Dec 13, 2022
5 Oct 21, 2022
53 Sep 7, 2022
7 Sep 8, 2022
22 Dec 11, 2022
6 Dec 7, 2021
7 May 29, 2022
3 Feb 10, 2022
11 Oct 04, 2022
185 Dec 29, 2022
1 Sep 11, 2022
0 Feb 24, 2022
32 Dec 01, 2022
2 Oct 29, 2021
769 Jan 03, 2023
20 Dec 20, 2022
0 Nov 27, 2021
3 Nov 06, 2021
1 Dec 07, 2021
3 Dec 28, 2022
5 Nov 08, 2021
33 Nov 22, 2022
37 Nov 02, 2022
16 Jan 07, 2023
1.5k Dec 31, 2022
4 May 01, 2022
4 Sep 24, 2021