Python library for SciDB streaming
Project description
Requirements
SciDB 16.9 or newer
Apache Arrow 0.6.0 or newer.
Python 2.7.x, 3.4.x, 3.5.x, 3.6.x or newer.
Required Python packages:
dill feather-format pandas
Note
Apache Arrow versions older than 0.8.0 contain a bug which might affect Stream users. The bug manifests on chunks of more than 128 records with null-able values. For more details, see the full bug description here. This bug has been fixed in Apache Arrow version 0.8.0.
Installation
Install latest release:
pip install scidb-strm
Install development version from GitHub:
pip install git+http://github.com/paradigm4/stream.git#subdirectory=py_pkg
The Python library needs to be installed on the SciDB server. The library needs to be installed on the client as well, if Python code is to be send from the client to the server.
SciDB-Strm Python API and Examples
Once installed the SciDB-Strm Python library can be imported with import scidbstrm. The library provides a high and low-level access to the SciDB stream operator as well as the ability to send Python code to the SciDB server.
High-level access is provided by the function map:
- map(map_fun, finalize_fun=None)
Read SciDB chunks. For each chunk, call map_fun and stream its result back to SciDB. If finalize_fun is provided, call it after all the chunks have been processed.
See 0-iquery.txt for a succinct example using the map function.
See 1-map-finalize.py for an example using the map function. The Python script has to be copied onto the SciDB instance.
Python code can be send to the SciDB server for execution using the pack_func and read_func functions:
- pack_func(func)
Serialize Python function for use as upload_data in input or load operators.
- read_func()
Read and de-serialize function from SciDB.
See 2-pack-func.py for an example of using the pack_func and read_func functions.
Low-level access is provided by the read and write functions:
- read()
Read a data chunk from SciDB. Returns a Pandas DataFrame or None.
- write(df=None)
Write a data chunk to SciDB.
See 3-read-write.py for an example using the read and write functions. The Python script has to be copied onto the SciDB instance.
A convenience invocation of the Python interpreter is provided in python_map variable and it is set to:
python -uc "import scidbstrm; scidbstrm.map(scidbstrm.read_func())"
Finally, see 4-machine-learning.py for a more complex example of going throught the steps of using machine larning (preprocessing, training, and prediction).
Debugging Python Code
When debugging Python code executed as part of the stream operator do not use the print function. The stream operator communicates with the Python process using stdout. The print function writes output to stdout. So, using the print function would interfere with the inter-process communication.
Instead, write debug output to stderr using the write function. For example:
import sys x = [1, 2, 3] sys.stderr.write("{}\n".format(x))
The output is written in the scidb-stderr.log files of each instance, for example:
/opt/scidb/18.1/DB-scidb/0/0/scidb-stderr.log /opt/scidb/18.1/DB-scidb/0/1/scidb-stderr.log
If using SciDB 18.1 installed in the default location and configured with one server and two instances.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for scidb_strm-18.1.1-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | b2079ac8f4a650fd79dd11399f3905670490f9b448f1793b7987a2d306e3e9ec |
|
MD5 | 3915ddc8d45bcf8879934c4dbf2f9867 |
|
BLAKE2b-256 | 71eca815f9b1cb55d2635b4dd70239458b6998813707f6d486bf1a5319060109 |