pradykaushik / Long-Short-Term-Memory-NNET-Tool

A wrapper around Keras Recurrent Layers to help build and test various configurations for Long Short Term Memory Neural Networks.

Command to run CLI

python run.py --infile <input filename> --outfile <output filename> --learnrate <(0.0,1.0]>

Other command-line options are given below.

Configuration parameters as command-line options

• Specify the learning rate (default value is 0.01) -- *required.

   --learnrate
  

  Determines the step size of the weight update and is critical to the convergence of the algorithm to the global minima of the objective function. A very high learning rate might end up missing certain local minima. A very low learning will end up slowing down the learning process.

• Specify the number of layers in the neural network.

   --nlayers
  

  Number of layers depends on the number of features that need to be extracted from the data.

• Specify the dimensions of each layer in the neural network.

   --layerdim
  

  Number of computation units in each layer.

• Specify the fraction of RUnits that are dropped out (value is in range [0.0, 1.0).

   --dropout_fraction_ru
  

  Drop a fraction of recurrent connections that remember the past results. Dropping RUnits would reduce exposure to data and hence, prevent overfitting.

• Specify the fraction of input units that are dropped out (value is in range [0.0, 1.0).

   --dropout_fraction_rw
  

  Drop fraction of inputs from passing to the next layer in the network. This helps in being more robust to noise.

• Specify the optimizer.

   --optimizer
  

  There are currently three optimizers (Adam, SGD and RMSprop).

• Specify the momentum.

   --momentum
  

  Weighted sum of past gradients that is used to accelerate learning and provide direction to the optimiser.

• Specify the training percent (The value is in range (0.0, 1.0]).

   --trainpct
  

  Percent of data to be used for training. The remaining would be used for testing and evaluation.

• Specify the error metric.

   --errmetric
  

• Specify the number of epochs.

   --epoch
  

  Number of passes through the whole training data.

Command-line options

• Specify input filename (.csv) -- *required.

   --infile
  

  The input file currently needs to contain 1-dimensional data.

• Specify output filename -- *required.

   --outfile
  

• Specify a config JSON file as input.

   --config
  

  Can use this to provide a file containing a JSON with appropriate parameters as the configuration to run the neural network. If config file provided, then all configuration parameters specified (those specified above) on the command-line would be ignored. Example config JSON is shown below,

  ```
  {  
    "n_layers": 4,  
    "dropout_fraction_ru": 0.1,  
    "dropout_fraction_rw": 0.1,  
    "layer_dimensions": [1, 60, 60, 1],  
    "optimizer": "adam",  
    "learning_rate": 0.001,  
    "momentum": 0.1,  
    "training_percent": 0.5,  
    "err_metric": "mean_squared_error",  
    "epoch": 10  
  }
  ```  
  

• Specify log filename (default logfile is _log).

   --logfile
  

• Append the run configuration to the logfile.

   --append
  

Command to run GUI

python gui.py

GUI Guide -- Configuration Input

GUI Guide -- Results