Timothy Yu

Archived notebooks are not "current" to the latest commit - usually anything in the archived directory has been implemented in python under `wsae_lstm`. My general development process uses jupyter notebooks...

@mg64ve 1. First the raw data is cleaned and then split into `train-validate-test` intervals: - https://github.com/timothyyu/wsae-lstm/blob/master/wsae_lstm/dataset/clean_dataset.py - https://github.com/timothyyu/wsae-lstm/blob/master/wsae_lstm/dataset/split_dataset.py ``` #print(dict_dataframes_index.keys()) # [index data][period 1-24][train/validate/test] # Train [1], Validate [2], Test...

Functions that are used to clean and split the dataset are in `wsae-lstm/utils.py`: https://github.com/timothyyu/wsae-lstm/blob/master/wsae_lstm/utils.py The function used to generate the report output in the `reports` folder for the `train-validate-test` split...

The raw data in the `data/raw` folder is straight from the source - it is the dataset that the authors of the WSAE-LSTM model journal/paper link and use themselves. Specifically,...

Interesting - scaling the indicators separately from the the OHLC is something I'm going to look into once I'm further along constructing the rest of the model. Additionally, I'm almost...

This is possible by saving the scaling parameters:  However, the same is not true for the denoise with the wavelet transform if the values for `sigma` are different: ```def...

> @timothyyu I don't think we need to be concerned about the reverse process. > ... > I can't really understand what I should expect after SAE. Should I expect...

@mg64ve also see https://github.com/timothyyu/wsae-lstm/issues/6 There are potential issues with how `sigma` and `uthresh` values are used for the wavelet transform that I am looking into

**Partial/incomplete answer to your question about the reverse process:** If the LSTMs are trained on scaled OHLC data, then the predictions will be scaled. If process is not reversible (even...

see comment on #9: https://github.com/timothyyu/wsae-lstm/issues/9#issuecomment-511061074