config.toml

# Model Config File
# This gets copied to model directory when training is started

[database]
train = './data/train_data.csv'
test = './data/test_data.csv'

[data_stats]
lmin = 0.0          # load minimum
lmax = 600.0        # load maximum (generation capacity)
tmin = 0.0          # temperature minimum
tmax = 120.0        # temperature maximum
yearStart = 1900.0  # counting year from 1900
yearEnd = 2100.0    # world ends in 2100

[model_arch]
modelArch = 'LSTM01'               # model architecture to use from modelLib.py
modelDir = './models'              # where models are saved
modelName = 'LSTM01_001'           # name to save model with

[model_params]      # params particular to an architecture (define models in modelLib.py)
seqLength = 360     # number of hours, 360 = 24 * 15 days
stepSize = 120      # number of hours offset between each sequence
inputDim = 7                          
outputDim = 1
denseUnits = [1024,1024,1024,1024]                      
denseActivation = ['prelu', 'prelu', 'prelu', 'prelu']
lstmUnits = [256,1]
lstmActivation = ['sigmoid','sigmoid']
batchnorm = true
dropout = 0.5

[training_params]
epochs = 100
batchSize = 512        
patience = 40           # number of epochs to wait before stopping if no change in loss
learningRate = 0.001
optimizer = 'adam'      # 'sgd' / 'adam' / 'adagrad' / 'adadelta'/ 'rmsprop' 
lossFunc = 'mse'        # 'mse' / 'mae' / 'mape'
metricFuncs = []
saveAllWeights = false