shunyaoshih / tpa-lstm Goto Github PK
View Code? Open in Web Editor NEWTemporal Pattern Attention for Multivariate Time Series Forecasting
Temporal Pattern Attention for Multivariate Time Series Forecasting
Hi! I'm interested in extracting the attention weights on each variable (I have 75 different variables) to determine the relative importance of each variable in predicting the output. How would I go about extracting these weights?
Can this be implemented with Keras?
AttributeError: module 'tensorflow.python.ops.rnn_cell_impl' has no attribute '_like_rnncell'
Traceback (most recent call last):
File "main.py", line 37, in
main()
File "main.py", line 16, in main
graph, model, data_generator = create_graph(para)
File "D:\study\keyan\paper\TPA-LSTM-master\lib\model_utils.py", line 32, in create_graph
model = PolyRNN(para, data_generator)
File "D:\study\keyan\paper\TPA-LSTM-master\lib\model.py", line 14, in init
self._build_graph()
File "D:\study\keyan\paper\TPA-LSTM-master\lib\model.py", line 26, in _build_graph
self.rnn_inputs, self.rnn_inputs_len, self.target_outputs = self.data_generator.inputs(
AttributeError: 'NoneType' object has no attribute 'inputs'
In the "_preprocess" function of the "TimeSeriesDataGenerator" class, why are there two types of normalization methods (specifically, the minmax normalization method for the "electricity" dataset; the other normalization method for the other time series datasets)?
Is the second normalization method reasonable? Because it involves the other variables' information when scaling a single variable.
when I change the data set to electricity and train, it will report a error by the assert self.para.highway == self.para.attention_len
, but when I delete this line, it will run but report the input can not be empty?
I don't understand the relationship between the parameter highway
and attention_len
, could you please give me some help? thanks a lot.
数据怎么换成自己数据?
when runs the train.py till lines 29
[loss, global_step, _] = sess.run(
fetches=[model.loss, model.global_step, model.update])
it shows error and turns to except tf.errors.OutOfRangeError:
why?
Hi! I find the usage of "call" function should be like "cell.call(inputs, state)", but in the "TemporalPatternAttentionCellWrapper" class, the "call" function is not used, I don't understand the code well. Could anyone teach me about this? Thanks a lot!
I can't get musedata from docs.google.com by url. how to solve it ?
Tried many ways, but logger only shows the code was run by CPU
May the code "new_state = (new_state, new_attns, new_attn_states)" should be written to "new_state = (new_state, output, new_attn_states)"?
Vice nice project!
But I'm confused with:
Why use CNN with the output of RNN, instead of original features to extract temporal
information for each variables? Maybe you do some experiments, but forget to mention it in paper?
With different batch, the maximum length of temporal series may be different. How can we deal this issue, if we use static kernel size?
Cheers!
Hi, we are a group of students from the University of Toronto. We read the research paper and really like the model. Our motivation for this adaptation stems from the shift in TensorFlow versions from 1.x to 2.x, which posed compatibility challenges for the original model. This project aims to make the model more accessible and maintainable by leveraging the flexibility and user-friendliness of PyTorch.
Recreate the Original Model: Faithfully adapt the model's architecture and functionality from TensorFlow to PyTorch.
Community Collaboration: Encourage contributions and feedback from the community to improve and evolve the model.
Installation: Instructions on setting up the environment and installing necessary dependencies.
Model Architecture: Detailed explanation of the model's architecture, including differences from the original version, if any.
Training and Evaluation: Step-by-step guide on how to train and evaluate the model using provided datasets or custom data.
Contributing: Guidelines for contributing to the project, including coding standards, submitting pull requests, and reporting issues.
Original Authors: Recognition of the authors of the original research paper and model.
Community Contributors: @shunyaoshih @Daikon-Sun
I was deploying your model on COLAB - with the repo copied, I ran the code below:
!python main.py --mode train
--attention_len 16
--batch_size 32
--data_set muse
--dropout 0.2
--learning_rate 1e-5
--model_dir ./models/model
--num_epochs 40
--num_layers 3
--num_units 338
Everything seems to be fine at that point, but it's stuck while training - I don't know why.T_T
https://colab.research.google.com/drive/1jaHTWg637wkD35C5LC5rRjej_G95L6tV?usp=sharing
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.