-
Notifications
You must be signed in to change notification settings - Fork 4
/
README_LSTM
36 lines (30 loc) · 2.05 KB
/
README_LSTM
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
### Readme for the LSTM classifier in the gesture recognition application
The LSTM module uses a neuronal network (NN) with LSTM blocks as hidden layer. For NN training is necessary.
The module provides both the classification with a trained network and the training itself. Additionally some
functions for better handling are provided.
### Features:
- create a neuronal network with different input dimension size, a hidden Layer of n LSTM Blocks and an output
Layer with different activation functions
- network can be trained with different training algorithms and different iterations and a trainig set.
- network can be validated against a test dataset
- network and datasets can be saved and loaded. The network is saved in xml format and can be modified manually.
Autosave and autoload is available
- almost all options are customizable through a personal config file.
- network can classify with different variants live events
- recognized gestures can be bound to a simulated key press of a keyboard (through config file)
- and other
### Usage:
Config parameters are explained in config/default.cfg. On first start a config/personal.cfg file is created. Changes can be made here.
An already trained network is provided and specified in default.cfg.
some commands for console:
u lstm - loads LSTM module
v - validates network against the testset and prints the error and a confmat
t - trains the network with the number of configured epochs
t <num> - rains the network with specifed epochs
p - prints the modules and weight of the network to console
c - start classification
# Use these commands with care as some states checks and parsing functions are not compatible to customnames
l <networkpath> - load network from this file (.xml will be appended automatically)
s <networkpath> - saves the current network to this path (.xml will be appended automatically)
l ds <datasetpath> - load dataset from this file (.data will be appended automatically)
s ds <datasetpath> - saves the current dataset to this path (.data will be appended automatically)