Problem

[xuyin218]

I'm sorry to disturb you again, I encountered some problems in the process of writing my code but I don't know how to solve it. So, I would like consult you: 1、Model.py: In "Localizer" class, "map_size” means maze size? 2、forward(self,map_in, obs_in, act_in): "map_in","obs_in","act_in"means? 3、I want to predict time-series, now do I need the function "step(self, map_in, obs_in, act_in, gt_pos, args)" ? Thank you very much! Best Regards.

[Yusufma03]

No worries.

1、Model.py: In "Localizer" class, "map_size” means maze size?

Yes.

2、forward(self,map_in, obs_in, act_in): "map_in","obs_in","act_in"means?

These are the input sequences, where map_in stands for the maze, obs_in is the local observation the robot receives, and act_in corresponds to the actions chosen.

3、I want to predict time-series, now do I need the function "step(self, map_in, obs_in, act_in, gt_pos, args)" ?

In my setup, yes.

However, if you are not familiar with deep learning and time series modeling yet, I would recommend you to start with some general time series modeling tutorials first. For example, the classic but useful ARIMA model. If you plan to do deep learning, this one explains how to code up an LSTM time series predictor step by step.

You can basically understand PF-RNNs as a variant of standard RNNs, e.g., LSTMs. Once you mastered the time series prediction using LSTM, then it becomes easy for you to use PF-RNNs.

[xuyin218]

No worries.

1、Model.py: In "Localizer" class, "map_size” means maze size?

Yes.

2、forward(self,map_in, obs_in, act_in): "map_in","obs_in","act_in"means?

These are the input sequences, where map_in stands for the maze, obs_in is the local observation the robot receives, and act_in corresponds to the actions chosen.

3、I want to predict time-series, now do I need the function "step(self, map_in, obs_in, act_in, gt_pos, args)" ?

In my setup, yes.

However, if you are not familiar with deep learning and time series modeling yet, I would recommend you to start with some general time series modeling tutorials first. For example, the classic but useful ARIMA model. If you plan to do deep learning, this one explains how to code up an LSTM time series predictor step by step.

You can basically understand PF-RNNs as a variant of standard RNNs, e.g., LSTMs. Once you mastered the time series prediction using LSTM, then it becomes easy for you to use PF-RNNs.

Thank you for your conscientious reply! Your advices are so useful! I think I must study the LSTM firstly, because I cannot understand some parts in your code. In you paper(AAAI2020), you said the PF-RNN model can predict stock price. Are you achieving it using python? how is the prediction accuracy? (Table 2 in you paper?) If you are willing to communicating with me, you can email to me.(xuyin218@mail.ustc.edu.cn)

[Yusufma03]

Are you achieving it using python?

Yes. I used exactly my current PF-RNN implementation provided here. However, the training pipeline is different from the robot localization. Currently, I'm not considering to release other experiments.

how is the prediction accuracy? (Table 2 in you paper?)

The prediction loss is given in the Table 1, NASDAQ entry.

[xuyin218]

Are you achieving it using python?

Yes. I used exactly my current PF-RNN implementation provided here. However, the training pipeline is different from the robot localization. Currently, I'm not considering to release other experiments.

how is the prediction accuracy? (Table 2 in you paper?)

The prediction loss is given in the Table 1, NASDAQ entry.

Thank you for sharing the code of the robot localization, otherwise I cannot know how to start. I can understand because writing codes is so hard!!! I will try to predict....

[Yusufma03]

No worries. Issue closed.