Eto Labs Python SDK
This is the python SDK for Eto, the AI focused data platform for teams bringing AI models to production. The python SDK makes it easy to integrate Eto's features into your AI training and analysis workflow.
Installation
The Eto python SDK is available on PyPI and can be installed via Pip:
pip install etosdk
Eto SDK is compatible with Python 3.7+
Setup
Before using the SDK for the first time, you must configure it with your Eto API url and the API token.
import eto
eto.configure(url='<eto-api-url>', token='<api-token>')
The above configuration function creates a configuration file under $XDG_CONFIG_HOME/eto/eto.conf, which is usually ~/.config/eto/eto.conf.
Ingesting data
To create an ingestion job to convert raw data in Coco format and create a new dataset:
import eto
job = eto.ingest_coco('<dataset_name>',
{'image_dir': '<path/to/images>',
'annotations': '<path/to/annotations>',
'extras': {'key': 'value'}})
The ingestion job will run asynchronously server-side and convert the data to Rikai (parquet) format. Once complete, you should be able to see it in the data registry:
import eto
eto.list_datasets() # list all datasets
eto.get_dataset('<dataset_name>') # get information about a single dataset
Analysis
Accessing a particular dataset is easy via Pandas:
import eto
import pandas as pd
df = pd.read_eto('<dataset_name>') # Eto SDK adds a pandas extension
Training
To train a pytorch model, you can use the Dataset/DataLoader classes in Rikai:
import eto
from rikai.torch.vision import Dataset
dataset = Dataset('<dataset_name>') # Eto SDK adds an extension to Rikai to resolve dataset references
for next_record in dataset:
# training loop
pass
A plain pytorch dataloader is also available from rikai.torch.data.DataLoader.
Local Spark configuration
For now, the Eto SDK relies on PySpark locally to read some of the custom Rikai types like annotations. While PySpark should be automatically installed as a transitive dependency, you may find that you need to change the Spark configurations to suit your local setup.
Your $SPARK_HOME/conf/spark-defaults.conf file should look something like the following:
spark.sql.extensions ai.eto.rikai.sql.spark.RikaiSparkSessionExtensions
spark.jars.packages ai.eto:rikai_2.12:0.0.13,org.apache.hadoop:hadoop-aws:3.2.0
# AWS
spark.executor.extraJavaOptions -Dcom.amazonaws.services.s3.enableV4=true -Dio.netty.tryReflectionSetAccessible=true
spark.driver.extraJavaOptions -Dcom.amazonaws.services.s3.enableV4=true -Dio.netty.tryReflectionSetAccessible=true
spark.hadoop.mapreduce.fileoutputcommitter.algorithm.version 2
spark.hadoop.com.amazonaws.services.s3.enableV4 true
fs.AbstractFileSystem.s3a.impl org.apache.hadoop.fs.s3a.S3A
fs.s3a.impl org.apache.hadoop.fs.s3a.S3AFileSystem
fs.s3a.aws.credentials.provider com.amazonaws.auth.InstanceProfileCredentialsProvider,com.amazonaws.auth.DefaultAWSCredentialsProviderChain
8 Oct 31, 2022
1 Jan 24, 2022
6 Mar 11, 2022
1 Oct 13, 2021
16 Sep 24, 2022
7.8k Jan 8, 2023
2.7k Jan 7, 2023
371 Dec 29, 2022
828 Jan 5, 2023
5 Sep 20, 2022
6 Feb 02, 2022
2 Oct 20, 2021
109 Dec 30, 2022
1 Jan 31, 2022
3 Mar 08, 2022
1 May 29, 2022
4 Nov 19, 2022
92 Apr 06, 2022
426 Jan 04, 2023
16 Oct 25, 2022
2.9k Jan 08, 2023
6 Jul 23, 2022
25 Oct 07, 2021
1 Nov 17, 2022
3 Mar 24, 2022
484 Jan 01, 2023
467 Jan 08, 2023
14 Nov 19, 2021
7 Apr 24, 2022