Created by: shboy

import sys, os

import paddle.v2 as paddle
import time


from STResNet import STResNet




def main():
    datadim_c = 6 * 234 * 236 
    datadim_p = 2 * 234 * 236 
    datadim_t = 2 * 234 * 236 
    datadim_ex = 31
    
    resdim = 2 * 234 * 236 
    
    if len(sys.argv) == 1:
        print('self def errror: input fomat error')
        print('need to specify  residual units')
        sys.exit(-1)
    else:
        nb_residual_unit = int(sys.argv[1])  # number of residual units

    # PaddlePaddle init
    paddle.init(use_gpu=True)

    input_c = paddle.layer.data(
        name="input_c", type=paddle.data_type.dense_vector(datadim_c), width=234, height=236)
    input_p = paddle.layer.data(
        name="input_p", type=paddle.data_type.dense_vector(datadim_p), width=234, height=236)
    input_t = paddle.layer.data(
        name="input_t", type=paddle.data_type.dense_vector(datadim_t), width=234, height=236)
    input_ex = paddle.layer.data(
        name="input_ex", type=paddle.data_type.dense_vector(datadim_ex))

    # Add neural network config
    # option 1. resnet
    main_output = STResNet(input_c, input_p, input_t, input_ex, external_dim=datadim_ex, nb_residual_unit=nb_residual_unit)

    def infer_data():
        def reader():
            #import numpy as np
            with open('/work/shenhao/test_data/event_day=2017010406070912_shuffle_test/2017010406070912', 'r') as f:
                line = f.readline().strip(' ').split()
                X_C = line[1: 331345]
                #X_C = np.asarray(X_C)
                X_P = line[331345: 331345+110448]
                #X_P = np.asarray(X_P)
                X_T = line[441793: 441793+110448]
                #X_T = np.asarray(X_T)
                X_E = line[552241: 552241+31]
                #X_E = np.asarray(X_E)
                return X_C, X_P, X_T, X_E
        return reader
    para_path = '/work/shenhao/paddle/STResNet_network_ok/params_pass_99.tar'

    with open(para_path, "r") as f:
        parameters = paddle.parameters.Parameters.from_tar(f)
    img = paddle.infer(
        output_layer=main_output,
        parameters=parameters,
        input=infer_data(),
        feeding={'input_c': 0,
               'input_p': 1,
               'input_t': 2,
               'input_ex': 3})
    print img

if __name__ == '__main__':
    ts_batch = time.time()
    ts_pass = time.time()
    main()