Created by: wfeagle

神经网络的最后一层softmax，用于预测时速度比较慢，能否复用训练的模型参数，然后将最后一层softmax换成线性的，其实就是想去除softmax的归一化因子。 尝试过，发现最后一层的参数不认，报如下错误： Traceback (most recent call last): File "infer.py", line 186, in main(args) File "infer.py", line 177, in main args.model_path) File "infer.py", line 112, in infer_net output_layer=pred_layer, parameters=parameters) File "/home/disk1/wangfang20/software/python27-gcc482/lib/python2.7/site-packages/paddle/v2/inference.py", line 50, in init val.copyFromNumpyArray(parameters.get(name).flatten()) File "/home/disk1/wangfang20/software/python27-gcc482/lib/python2.7/site-packages/paddle/v2/parameters.py", line 242, in get return self.getitem(key=parameter_name) File "/home/disk1/wangfang20/software/python27-gcc482/lib/python2.7/site-packages/paddle/v2/parameters.py", line 186, in getitem return self.__getter_inner(key, api.PARAMETER_VALUE) File "/home/disk1/wangfang20/software/python27-gcc482/lib/python2.7/site-packages/paddle/v2/parameters.py", line 153, in __getter_inner shape = self.get_shape(key) File "/home/disk1/wangfang20/software/python27-gcc482/lib/python2.7/site-packages/paddle/v2/parameters.py", line 200, in get_shape raise ValueError("No such parameter %s" % key) ValueError: No such parameter __fc_layer_0.w0 PS：我不关心训练速度，只想在预测时加速，因为想放到线上使用