All Projects → didi → Dlflow

didi / Dlflow

Licence: apache-2.0
DLFlow is a deep learning framework.

Programming Languages

python
139335 projects - #7 most used programming language

Projects that are alternatives of or similar to Dlflow

Wb color augmenter
WB color augmenter improves the accuracy of image classification and image semantic segmentation methods by emulating different WB effects (ICCV 2019) [Python & Matlab].
Stars: ✭ 89 (-15.24%)
Mutual labels:  deeplearning
Ngsim env
Learning human driver models from NGSIM data with imitation learning.
Stars: ✭ 96 (-8.57%)
Mutual labels:  deeplearning
Mit Deep Learning Book Pdf
MIT Deep Learning Book in PDF format (complete and parts) by Ian Goodfellow, Yoshua Bengio and Aaron Courville
Stars: ✭ 9,859 (+9289.52%)
Mutual labels:  deeplearning
Lipreading Densenet3d
DenseNet3D Model In "LRW-1000: A Naturally-Distributed Large-Scale Benchmark for Lip Reading in the Wild", https://arxiv.org/abs/1810.06990
Stars: ✭ 91 (-13.33%)
Mutual labels:  deeplearning
360sd Net
Pytorch implementation of ICRA 2020 paper "360° Stereo Depth Estimation with Learnable Cost Volume"
Stars: ✭ 94 (-10.48%)
Mutual labels:  deeplearning
Exposure correction
Reference code for the paper "Learning Multi-Scale Photo Exposure Correction", CVPR 2021.
Stars: ✭ 98 (-6.67%)
Mutual labels:  deeplearning
Novel Deep Learning Model For Traffic Sign Detection Using Capsule Networks
capsule networks that achieves outstanding performance on the German traffic sign dataset
Stars: ✭ 88 (-16.19%)
Mutual labels:  deeplearning
Spectralnormalizationkeras
Spectral Normalization for Keras Dense and Convolution Layers
Stars: ✭ 100 (-4.76%)
Mutual labels:  deeplearning
Bdrar
Code for the ECCV 2018 paper "Bidirectional Feature Pyramid Network with Recurrent Attention Residual Modules for Shadow Detection"
Stars: ✭ 95 (-9.52%)
Mutual labels:  deeplearning
Mad Twinnet
The code for the MaD TwinNet. Demo page:
Stars: ✭ 99 (-5.71%)
Mutual labels:  deeplearning
Micromlp
A micro neural network multilayer perceptron for MicroPython (used on ESP32 and Pycom modules)
Stars: ✭ 92 (-12.38%)
Mutual labels:  deeplearning
Classification Of Hyperspectral Image
Classification of the Hyperspectral Image Indian Pines with Convolutional Neural Network
Stars: ✭ 93 (-11.43%)
Mutual labels:  deeplearning
Intra Bag And Inter Bag Attentions
Code for NAACL 2019 paper: Distant Supervision Relation Extraction with Intra-Bag and Inter-Bag Attentions
Stars: ✭ 98 (-6.67%)
Mutual labels:  deeplearning
Pytorch Seq2seq
An open source framework for seq2seq models in PyTorch.
Stars: ✭ 1,297 (+1135.24%)
Mutual labels:  deeplearning
Androidtensorflowmachinelearningexample
Android TensorFlow MachineLearning Example (Building TensorFlow for Android)
Stars: ✭ 1,369 (+1203.81%)
Mutual labels:  deeplearning
Machine learning code
机器学习与深度学习算法示例
Stars: ✭ 88 (-16.19%)
Mutual labels:  deeplearning
The Nlp Pandect
A comprehensive reference for all topics related to Natural Language Processing
Stars: ✭ 1,349 (+1184.76%)
Mutual labels:  deeplearning
Matex
Machine Learning Toolkit for Extreme Scale (MaTEx)
Stars: ✭ 104 (-0.95%)
Mutual labels:  deeplearning
Sert
Semantic Entity Retrieval Toolkit
Stars: ✭ 100 (-4.76%)
Mutual labels:  deeplearning
Har Keras Cnn
Human Activity Recognition (HAR) with 1D Convolutional Neural Network in Python and Keras
Stars: ✭ 97 (-7.62%)
Mutual labels:  deeplearning

contributions license python

DLFlow - A Deep Learning WorkFlow

DLFlow概述

DLFlow是一套深度学习pipeline,它结合了Spark的大规模特征处理能力和Tensorflow模型构建能力。利用DLFlow可以快速处理原始特征、训练模型并进行大规模分布式预测,十分适合离线环境下的生产任务。利用DLFlow,用户只需专注于模型开发,而无需关心原始特征处理、pipeline构建、生产部署等工作。

功能支持

配置驱动: DLFlow通过配置驱动,修改配置可以快速更换特征、模型超参数、任务流程等等,极大提高工作效率。

模块化结构: 任务和模型以插件形式存在,便于使用和开发,用户可以可以轻地将自定义任务和模型注册到框架内使用。

任务自组织: 通过内置的Workflow框架,根据任务的产出标记自动解决任务依赖,轻松构建深度学习pipeline。

最佳实践: 融入滴滴用户画像团队深度学习离线任务的最佳实践,有效应对离线生产中的多种问题。将Tensorflow和Spark进行合理结合,更适合离线深度学习任务。

快速开始

环境准备

首先请确保环境中已经安装和配置Hadoop和Spark,并设置好了基本的环境变量。

  • Tensorflow访问HDFS

为了能够使用让Tensorflow访问HDFS,需要确保如下环境变量生效:

# 确保libjvm.so被添加到LD_LIBRARY_PATH
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${JAVA_HOME}/jre/lib/amd64/server

# 确保hadoop jars被添加到CLASSPATH
export CLASSPATH=${CLASSPATH}:$(hadoop classpath --glob)

关于Tensorflow访问HDFS更多内容请参见 TensorFlow on Hadoop

  • Spark读写TFReocrds
# Clone tensorflow/ecosystem项目
git clone https://github.com/tensorflow/ecosystem.git

cd ecosystem/spark/spark-tensorflow-connector/

# 构建spark-tensorflow-connector
mvn versions:set -DnewVersion=1.14.0
mvn clean install

项目构建后生成 target/spark-tensorflow-connector_2.11-1.14.0.jar,后续需要确保该jar被添加到 spark.jars 中。 关于Spark读写TFRecoreds更多内容请参见 spark-tensorflow-connector

安装

通过pip安装:

pip install dlflow

通过源代码安装:

git clone  https://github.com/didi/dlflow.git
cd dlflow
python setup.py install

使用

  • 配置文件

运行配置可参考 conf 目录中的配置。 关于配置详情请参考 配置说明

  • 以模块运行
python -m dlflow.main --config <CONFIGURATION FILE>.conf
  • 以脚本运行

确保python环境的 bin 目录已经被添加到环境变量 PATH

export PATH=$PATH:/usr/local/python/bin

之后通过如下命令运行

dlflow --config <CONFIGURATION FILE>.conf

更详细的使用参见 使用说明

预定义任务

预定义任务 描述
Merge 特征融合任务,请参见 特征融合
Encode 解析原始特征,对特征进行编码和预处理,生成能够直接输入模型的特征
Train 模型训练任务
Evaluate 模型评估任务
Predict 模型预测任务,使用Spark进行分布式预测,具备处理大规模数据能力

手册目录

技术方案

DLFlow整体架构

整体架构

DLFLow pipeline

Pipeline

Contributing

欢迎使用并参与到本项目的建设中,详细内容请参见 Contribution Guide

License

DLFlow 基于Apache-2.0协议进行分发和使用,更多信息参见 LICENSE

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].