预测示例 (R)¶

本章节包含2部分内容：(1) 运行 R 示例程序；(2) R 预测程序开发说明。

运行 R 示例程序¶

1. 安装 R 预测环境¶

方法1: Paddle Inference 的 R 语言预测依赖 Paddle Python环境，请先根据官方主页-快速安装页面进行自行安装或编译，当前支持 pip/conda 安装，docker镜像以及源码编译等多种方式来准备 Paddle Inference Python 环境。之后需要安装 R 运行paddle预测所需要的库

Rscript -e 'install.packages("reticulate", repos="https://cran.rstudio.com")'

方法2: 将 Paddle/r/Dockerfile 下载到本地，使用以下命令构建 Docker 镜像，启动 Docker 容器：

# 构建 Docker 镜像
docker build -t paddle-rapi:latest .

# 启动 Docker 容器
docker run --rm -it paddle-rapi:latest bash

2. 准备预测部署模型¶

下载 ResNet50 模型后解压，得到 Paddle 预测格式的模型，位于文件夹 ResNet50 下。如需查看模型结构，可将 inference.pdmodel 文件重命名为 __model__，然后通过模型可视化工具 Netron 打开。

wget https://paddle-inference-dist.bj.bcebos.com/Paddle-Inference-Demo/resnet50.tgz
tar zxf resnet50.tgz

# 获得模型目录即文件如下
resnet50/
├── inference.pdmodel
├── inference.pdiparams.info
└── inference.pdiparams

3. 准备预测部署程序¶

将以下代码保存为 r_demo.r 文件，并添加可执行权限：

#!/usr/bin/env Rscript

library(reticulate) # call Python library
use_python("/opt/python3.7/bin/python")

np <- import("numpy")
paddle_infer <- import("paddle.inference")

predict_run_resnet50 <- function() {
    # 创建 config
    config <- paddle_infer$Config("resnet50/inference.pdmodel", "resnet50/inference.pdiparams")
    
    # 根据 config 创建 predictor
    predictor <- paddle_infer$create_predictor(config)

    # 获取输入的名称
    input_names <- predictor$get_input_names()
    input_handle <- predictor$get_input_handle(input_names[1])

    # 设置输入
    input_data <- np$random$randn(as.integer(1 * 3 * 318 * 318))
    input_data <- np_array(input_data, dtype="float32")$reshape(as.integer(c(1, 3, 318, 318)))
    input_handle$reshape(as.integer(c(1, 3, 318, 318)))
    input_handle$copy_from_cpu(input_data)

    # 运行predictor
    predictor$run()

    # 获取输出
    output_names <- predictor$get_output_names()
    output_handle <- predictor$get_output_handle(output_names[1])
    output_data <- output_handle$copy_to_cpu()
    output_data <- np_array(output_data)$reshape(as.integer(-1))
    print(paste0("Output data size is: ", output_data$size))
    print(paste0("Output data shape is: ", output_data$shape))
}

if (!interactive()) {
    predict_run_resnet50()
}

4. 执行预测程序¶

# 将本章节第2步中下载的模型文件夹移动到当前目录
./r_demo.r

成功执行之后，得到的预测输出结果如下：

# 程序输出结果如下
grep: warning: GREP_OPTIONS is deprecated; please use an alias or script
W1215 10:48:45.627841 52293 analysis_predictor.cc:134] Profiler is activated, which might affect the performance
--- Running analysis [ir_graph_build_pass]
--- Running analysis [ir_graph_clean_pass]
--- Running analysis [ir_analysis_pass]
--- Running IR pass [simplify_with_basic_ops_pass]
--- Running IR pass [attention_lstm_fuse_pass]
--- Running IR pass [seqconv_eltadd_relu_fuse_pass]
--- Running IR pass [seqpool_cvm_concat_fuse_pass]
--- Running IR pass [mul_lstm_fuse_pass]
--- Running IR pass [fc_gru_fuse_pass]
--- Running IR pass [mul_gru_fuse_pass]
--- Running IR pass [seq_concat_fc_fuse_pass]
--- Running IR pass [fc_fuse_pass]
I1215 10:48:46.117144 52293 graph_pattern_detector.cc:100] ---  detected 1 subgraphs
--- Running IR pass [repeated_fc_relu_fuse_pass]
--- Running IR pass [squared_mat_sub_fuse_pass]
--- Running IR pass [conv_bn_fuse_pass]
--- Running IR pass [conv_eltwiseadd_bn_fuse_pass]
I1215 10:48:46.341869 52293 graph_pattern_detector.cc:100] ---  detected 53 subgraphs
--- Running IR pass [conv_transpose_bn_fuse_pass]
--- Running IR pass [conv_transpose_eltwiseadd_bn_fuse_pass]
--- Running IR pass [is_test_pass]
--- Running IR pass [runtime_context_cache_pass]
--- Running analysis [ir_params_sync_among_devices_pass]
--- Running analysis [adjust_cudnn_workspace_size_pass]
--- Running analysis [inference_op_replace_pass]
--- Running analysis [ir_graph_to_program_pass]
I1215 10:48:46.388198 52293 analysis_predictor.cc:537] ======= optimize end =======
I1215 10:48:46.388363 52293 naive_executor.cc:102] ---  skip [feed], feed -> data
I1215 10:48:46.389770 52293 naive_executor.cc:102] ---  skip [AddmmBackward190.fc.output.1.tmp_1], fetch -> fetch
[1] "Output data size is: 512"
[1] "Output data shape is: (512,)"

R 预测程序开发说明¶

使用 Paddle Inference 开发 R 预测程序仅需以下五个步骤：

(1) 在 R 中引入 Paddle Python 预测库

library(reticulate) # 调用Paddle
use_python("/opt/python3.7/bin/python")

np <- import("numpy")
paddle_infer <- import("paddle.inference")

(2) 创建配置对象，并根据需求配置，详细可参考 Python API 文档 - Config

# 创建 config，并设置预测模型路径
config <- paddle_infer$Config("resnet50/inference.pdmodel", "resnet50/inference.pdiparams")

(3) 根据Config创建预测对象，详细可参考 Python API 文档 - Predictor

predictor <- paddle_infer$create_predictor(config)

(4) 设置模型输入 Tensor，详细可参考 Python API 文档 - Tensor

# 获取输入的名称
input_names <- predictor$get_input_names()
input_handle <- predictor$get_input_handle(input_names[1])

# 设置输入
input_data <- np$random$randn(as.integer(1 * 3 * 318 * 318))
input_data <- np_array(input_data, dtype="float32")$reshape(as.integer(c(1, 3, 318, 318)))
input_handle$reshape(as.integer(c(1, 3, 318, 318)))
input_handle$copy_from_cpu(input_data)

(5) 执行预测，详细可参考 Python API 文档 - Predictor

predictor$run()

(5) 获得预测结果，详细可参考 Python API 文档 - Tensor

output_names <- predictor$get_output_names()
output_handle <- predictor$get_output_handle(output_names[1])
output_data <- output_handle$copy_to_cpu()
output_data <- np_array(output_data)$reshape(as.integer(-1)) # numpy.ndarray类型