+2 votes
5 views
in AI and Deep Learning by (330 points)
I saw a few examples of a neural network but they work for a fixed set of inputs. How can I deal with variable data length like string etc? Is there any way to encode variable data length into a fixed set of inputs and still get the generalized property of neural networks?

2 Answers

+3 votes
by (10.9k points)
edited by
It is not advised to just extract some data and feeding it to the network which most people do. In practical, success and performance of a neural network depends on the Pre-processing and choosing of right features. The ANN is made for fixed feature vector length so you won't be able to use it.But you may use other classifiers such as random forest which uses a small no. of random features and creates hundreds of decision trees or can also try other solutions like using the Dynamic Time Wrapping or the interpolation.
0 votes
by (31.3k points)

The main concept of the neural network is that we extract features from our data and then feed them to the input of the network. The feature selection and pre-processing decide the success of your network. It needs practice upon various datasets to pick the best features.

"Garbage in, Garbage out"

The above statement fits well with your question. If we feed un-processed data then the results can’t be good.

Hope this answer helps.

...