def load_data(filename):
"""Loads CSV for splitting into training and testing data.
Parameters:
filename : the filename of the file to load.
Returns:
Two lists, each corresponding to training and testing data.
"""
# get holdout ratio
train_percent = par.get_holdout()
# load into pandas dataframe
df = pd.read_csv(filename, header=None, dtype=float)
# normalize the data
for features in range(len(df.columns) - 1):
df[features] = (df[features] -
df[features].mean()) / df[features].std()
train = df.sample(frac=train_percent).fillna(0.00) # get training portion
test = df.drop(train.index).fillna(0.00) # remainder testing portion
return train.values.tolist(), test.values.tolist()
Returns:
The neuron activation based on the summed output.
"""
return z if z >= 0 else 0.01 * z
if __name__ == '__main__':
# if executed from automation script
if len(argv) == 3:
AUTO = bool(int(argv[2]))
else:
AUTO = False
MSE, TRP, TEP = [], [], [] # set up variables to store testing data
# load data to train and test network on
TRAIN, TEST = io.load_data(f'../data/{argv[1]}.csv', par.get_holdout())
# network-specific parameters
FEATURES = len(TRAIN[0][:-1]) # number of attributes of data
CLASSES = len({c[-1] for c in TRAIN + TEST}) # distinct classifications
HIDDEN_SIZE = par.get_hidden_size(argv[1])
DIMENSIONS = (HIDDEN_SIZE * (FEATURES + 1)) + (CLASSES * (HIDDEN_SIZE + 1))
EPOCHS, AXIS_RANGE = par.get_epochs(), par.get_rand_range()
# de-specific parameters
POP_SIZE = par.get_de_population_size()
CROSS_RATE, DIFF_WEIGHT = par.get_de_params(argv[1])
# run the de-nn
differential_evolution(DIMENSIONS, EPOCHS, POP_SIZE, AXIS_RANGE, \
CROSS_RATE, DIFF_WEIGHT)
if not AUTO:
io.plot_data(EPOCHS, MSE, TRP, TEP)
exit(0)