How does being bidirectional help in time-series, how is it possible for the past values to be influenced by future values?
Hello @Lakshmi_Narayana ,
Bidirectional LSTMs enables additional training by traversing the input data twice .
It process the given input sequence in both the directions - forward and backward directions.
The backward LSTM layer processes the input sequence from the end to the beginning.
The forward LSTM layer processes the input sequence from the beginning to the end.
Then concatenation of both output layers takes place to generate the final output .
This helps the model to capture both short-term and long-term dependencies in the time series. Therefore the network can learn both past and future values of time series.
Like this, Bidirectional LSTMs (BiLSTMs) optimize time-series forecasting .
WIth regards,
Nilosree Sengupta
Hi Nilosree,
Thank you for the explanation.
I understand that bidirectional helps in sentences as future words can determine the meaning of the current word but in a time series, I don’t understand why the future values are even being considered for each value, as I don’t see future values influencing the past values as in sentences.
Hello @Lakshmi_Narayana ,
You are most welcome!
For next word prediction you may not understand the functionality.
So I am giving you another example of stock market prediction.
Here it gets complex. Stock market follows a pattern that not only depends on news but also technicals-curves. We see patterns getting repeated several times periodically. Or price change following a pattern and also relativity in indices.
For example if a stock price falls, price goes down, demand increases-> people start buying → price again increases ->again demand falls after touching a respective high point.
So here we see the up and down.
Bidirectional LSTMs enable additional training by traversing the input data twice, left-to-right and right-to-left, allowing the model to learn from both directions.
Analysts studies the price levels. Predicts target price, how much it can fall, where is the support, how much it can go up, where is the resistance.
Unidirectional Lstms can’t capture such complicated stuff.
We need to study both directions- So Bidirectional LSTMs.
Hope this helps.
With regards,
Nilosree Sengupta