def evaluate_ind(individual):
"""
Fitness function.
Trains a randomly initiated ESN using the parameters in 'individual' and
the config file.
Returns touple with error metric (touple required by DEAP)
"""
parameters = paramhelper.get_parameters(individual)
# Run a few times to get the average error over several networks.
n_eval = optconfig['n_eval']
errors = np.empty((n_eval, ), dtype=float)
for i in range(n_eval):
_, errors[i] = esnet.run_from_config(Xtr, Ytr, Xval, Yval, parameters)
error = np.mean(errors)
# Do we have dimensionality reduction?
if parameters['n_dim'] is None:
return error,
else:
return error, float(
parameters['n_dim']) / parameters['n_internal_units']
def single_run(dummy):
"""
This function will be run by the workers.
"""
_,error = esnet.run_from_config(Xtr, Ytr, Xte, Yte, config)
return error
def single_run(dummy):
"""
This function will be run by the workers.
"""
Y_pred, error = esnet.run_from_config(Xtr, Ytr, Xte, Yte, config)
np.savetxt('Y_pred_prep', Y_pred)
return error
def single_run(dummy):
"""
This function will be run by the workers.
"""
predictions, error = esnet.run_from_config(Xtr, Ytr, Xte, Yte, config,
Yscaler)
predictions = Yscaler.inverse_transform(predictions)
allPredictions.append(predictions)
return error
def main():
averages = []
predictions_error = []
predictions = []
reals = []
startPoint = args.count
#For the first few predictions, use the last error as prediction
dataPath = '/home/minh/PycharmProjects/Ensemble/PythonESN/data/edgar_historical'
count = 0
"""with open(dataPath,'r') as f:
for line in f:
if count>0:
data = line.split(',')
total = 0
for i in range(5):
total += float(data[i])
averages.append(total/5)
reals.append(float(data[6]))
count+=1"""
"""count = 0
with open(dataPath,'r') as f:
for line in f:
if count>0:
if count>(startPoint+2):
break
else:
data = line.split(',')
total = 0
for i in range(5):
total += float(data[i])
predictions.append(float(data[6])-total/5)
count+=1
print('predictions:',predictions)"""
# Run in parallel and store result in a numpy array
X, Y = esnet.load_from_text(args.data)
Xtr, Ytr, _, _, Xte, Yte, Yscaler = esnet.generate_datasets(X[:startPoint], Y[:startPoint])
Yhat, error = esnet.run_from_config(Xtr, Ytr, Xte, Yte, config, Yscaler)
Yhat = np.ceil(Yscaler.inverse_transform(Yhat))
predictions_error.append(Yhat[len(Yhat) - 1][0])
"""for i in range(startPoint,len(X)):
Xtr, Ytr, _, _, Xte, Yte, Yscaler = esnet.generate_datasets(X[:i], Y[:i])
if i<800:
config['n_drop'] = int(i/8)
else:
config['n_drop'] = 100
Yhat, error = esnet.run_from_config(Xtr, Ytr, Xte, Yte, config)
Yhat = np.ceil(Yscaler.inverse_transform(Yhat))
#print('predictions:',Yhat)
#print('error:',error)
predictions_error.append(Yhat[len(Yhat)-1][0])
count+=1
if count%100==0:
print('predictions made:',count)"""
curPath = os.getcwd().split('/')
writePath = ''
for i in range(len(curPath)):
writePath += curPath[i] + '/'
configs = args.esnconfig.split('/')
writePath += 'predictions/predictions_' + configs[-1] + '_' + str(args.times)
"""with open(writePath, 'a') as f:
f.write(str(np.ceil(Yhat[len(Yhat) - 1][0]))+'\n')"""
with open (writePath, 'w') as f:
for i in range(len(Yhat)):
f.writelines(str(np.ceil(Yhat[i][0])) + '\n')