From 0a91acd93ecc8bafb170526bc070fbec7305db00 Mon Sep 17 00:00:00 2001 From: wuzewu Date: Tue, 2 Apr 2019 22:00:14 +0800 Subject: [PATCH] update README --- README.md | 37 ++++++++++++++++---------- docs/command_line_introduction.md | 23 +++++++++++++++++ docs/released_module_list.md | 10 ++++++++ docs/transfer_learning_turtorial.md | 40 +++++++++++++++++++++++++++++ 4 files changed, 96 insertions(+), 14 deletions(-) create mode 100644 docs/command_line_introduction.md create mode 100644 docs/released_module_list.md create mode 100644 docs/transfer_learning_turtorial.md diff --git a/README.md b/README.md index 520b84a4..88658ae0 100644 --- a/README.md +++ b/README.md @@ -1,20 +1,29 @@ # PaddleHub +PaddleHub旨在为PaddlePaddle提供一个简明易用的预训练模型管理框架。 +使用PaddleHub，你可以： +1. 通过统一的方式，快速便捷的获取PaddlePaddle发布的预训练模型 +2. 利用PaddleHub提供的接口，对预训练模型进行Transfer learning +3. 以命令行或者python代码调用的方式，使用预训练模型进行预测 -[![Build Status](https://travis-ci.org/PaddlePaddle/PaddleHub.svg?branch=develop)](https://travis-ci.org/PaddlePaddle/PaddleHub) -[![License](https://img.shields.io/badge/license-Apache%202-blue.svg)](LICENSE) -[![Version](https://img.shields.io/github/release/PaddlePaddle/PaddleHub.svg)](https://github.com/PaddlePaddle/PaddleHub/releases) - -## 安装 +除此之外，我们还提供了预训练模型的本地管理机制（类似于pip），用户可以通过命令行来管理本地的预训练模型 +![图片](http://agroup-bos.cdn.bcebos.com/89dc20492e986c262d8e3957e516a8c509413ccc) +想了解PaddleHub已经发布的模型，请查看 +# 安装 +paddle hub直接通过pip进行安装（python3以上），使用如下命令来安装paddle hub ``` -pip install paddlehub - +pip install paddle_hub ``` - -## 答疑 - -欢迎您将问题和bug报告以[Github Issues](https://github.com/PaddlePaddle/PaddleHub/issues)的形式提交 - -## 版权和许可证 -PaddleHub由[Apache-2.0 license](LICENSE)提供 +# 快速体验 +通过下面的命令，来体验下paddle hub的魅力 +``` +#使用lac进行分词 +hub run lac --input_text "今天是个好日子" +#使用senta进行情感分析 +hub run senta --input_text "今天是个好日子" +``` +# 深入了解Paddle Hub +* 命令行功能 +* Transfer Learning +* API diff --git a/docs/command_line_introduction.md b/docs/command_line_introduction.md new file mode 100644 index 00000000..db2aa312 --- /dev/null +++ b/docs/command_line_introduction.md @@ -0,0 +1,23 @@ +# 命令行 +Paddle Hub为Module的管理和使用提供了命令行工具，目前命令行支持以下9个命令： + +* install +* uninstall +* show +* download +* search +* list +* run +* help +* version + +## install +install命令用于将 +## uninstall +## show +## download +## search +## list +## run +## help +## version diff --git a/docs/released_module_list.md b/docs/released_module_list.md new file mode 100644 index 00000000..39939238 --- /dev/null +++ b/docs/released_module_list.md @@ -0,0 +1,10 @@ +# 已发布模型列表 +|方向 | 模型 | 描述 | +|---|---|---| +|NLP | ERNIE | | +|NLP | BERT | | +|NLP | LAC | 中文词性分析工具，支持中文分词、词性标注、命名实体识别等三个任务，可以直接用命令行预测 | +|NLP | SENTA | 中文情感分析模型，将结果分为三个级别，2分表示正面评价，1分表示中性评价，0分表示负面评价，支持命令行预测 | +|CV | ResNet | 使用ImageNet-2012数据集训练的分类模型，提供了classification和feature_map两个签名，其中classification签名支持直接通过命令行预测，feature_map签名用于finetune | +|CV | MobileNet | 使用ImageNet-2012数据集训练的分类模型，提供了classification和feature_map两个签名，其中classification签名支持直接通过命令行预测，feature_map签名用于finetune | +|CV | SSD | 在PASCAL VOC数据集上训练的SSD模型，可以用于进行20个类别目标的检测和定位，支持命令行预测 | diff --git a/docs/transfer_learning_turtorial.md b/docs/transfer_learning_turtorial.md new file mode 100644 index 00000000..25320ba6 --- /dev/null +++ b/docs/transfer_learning_turtorial.md @@ -0,0 +1,40 @@ +# Transfer Learning +Transfer Learning是xxxx +更多关于Transfer Learning的知识，请参考 +## CV教程 +以猫狗分类为例子，我们可以快速的使用一个通过ImageNet训练过的ResNet进行finetune +```python +import paddle.fluid as fluid +import paddle_hub as hub + +resnet = hub.Module(key = "resnet_v2_50_imagenet") +input_dict, output_dict, program = resnet.context(sign_name = "feature_map") +img_mode, img_size, img_order = resnet.data_config() +reader = hub.ImageClassifierReader(mode = img_mode, shape = img_shape, order = img_order, dataset = hub.dataset.flowers(), batch_size = 32) +with fluid.program_guard(program): + img = input_dict["image"] + feature_map = output_dict["feature_map"] + label = fluid.layers.data(name = "label", shape = [1], dtype = "int64") + task = hub.DNNClassifier(input = feature_map, hidden_units = [10], acts = ["softmax"]) + +finetune_config = {"epochs" : 100} +hub.finetune_and_eval(task = task, reader = reader.train(), config = finetune_config) +``` +## NLP教程 +```python +import paddle.fluid as fluid +import paddle_hub as hub + +resnet = hub.Module(key = "resnet_v2_50_imagenet") +input_dict, output_dict, program = resnet.context(sign_name = "feature_map") +img_mode, img_size, img_order = resnet.data_config() +reader = hub.ImageClassifierReader(mode = img_mode, shape = img_shape, order = img_order, dataset = hub.dataset.flowers(), batch_size = 32) +with fluid.program_guard(program): + img = input_dict["image"] + feature_map = output_dict["feature_map"] + label = fluid.layers.data(name = "label", shape = [1], dtype = "int64") + task = hub.DNNClassifier(input = feature_map, hidden_units = [10], acts = ["softmax"]) + +finetune_config = {"epochs" : 100} +hub.finetune_and_eval(task = task, reader = reader.train(), config = finetune_config) +``` -- GitLab