tsgan_configuration.yml

fit : True                          #set to True when fitting/training, False for generating new data
sample_new_data : False             #set to False when fitting/training, True for generating new data
dataset:
    class : ts
    path  : ./data/                 #dir with data
    filename : RHHBY.csv            #input file
    datetime_col  : Date            # datetime column
    value_col : Close               # value column
    time_window : 30                # time window
workers : 0                         #number of data loading workers
batch_size : 16                     #input batch size
nz: 100                             #dimensionality of the latent vector z
epochs : 70                         #number of epochs to train for
lr : 0.0002                         #learning rate
cuda : True                         #enables cuda
netG_path :    '' #'checkpoints/netG_test_CNN_dis_LSTM_gen_1hourtw_deltaminmax_epoch_35.pth'                        #path to netG (to continue training)
netD_path :    '' #'checkpoints/netD_test_CNN_dis_LSTM_gen_1hourtw_deltaminmax_epoch_35.pth'                        #path to netD (to continue training)
outf : checkpoints                  #folder to save checkpoints
imf : images                        #folder to save images
logdir : log                        #logdir for tensorboard
run_tag : test_CNN_dis_LSTM_gen     #tags for the current run
checkpoint_every : 5                #number of epochs after which saving checkpoints
tensorboard_image_every : 5         #interval for displaying images on tensorboard 
delta_condition : True              #whether to use the mse loss for deltas
delta_lambda : 10                   #weight for the delta condition
alternate : True                    #whether to alternate between adversarial and mse loss in generator
dis_type : cnn                      #choices=['cnn','lstm'], architecture to be used for discriminator to use
gen_type : lstm                     #choices=['cnn','lstm'], architecture to be used for generator to use
#output_path : ./output
deltas : [0.1, 0.2, 0.3, 0.05]      #deltas for generation, if only one value a number=size of sequences (length = time_window) will be created, if multiple values a sequence per value will be created
size : 100                          #in case of generation with only one delta value size is the number of sequences generated, otherwise it is not used. If generation without delta it is the length of the sequence.
outfile : ./test_CNN_dis_LSTM_gen.npy