def run(res, trn):
esn = pyESN.ESN(n_inputs=1,
n_outputs=3,
n_reservoir=res,
spectral_radius=0.9,
sparsity=0.002,
random_state=42)
spin_off = 0
trainN = trn
testN = 2000
sol_matrix = np.zeros([testN, 3])
np.shape(sol_matrix)
esn.fit(np.ones(trainN - spin_off), dataT[spin_off:trainN, :])
prediction = esn.predict(np.ones(testN))
error_mat = prediction - dataT[trainN:trainN + testN]
horizon = 0
while (horizon < testN) and \
(abs(error_mat[horizon, 0]) < 0.5) and \
(abs(error_mat[horizon, 1]) < 0.5) and \
(abs(error_mat[horizon, 2]) < 0.5):
horizon += 1
return (np.sum(np.sum(np.square(error_mat[:horizon])))) / float(
np.sum(np.sum(np.square(dataT[trainN:trainN + horizon])))), horizon
def __init__(self, input_dim, output_dim=1):
self.input_dim = input_dim
self.ESN = pyESN.ESN(n_inputs=input_dim,
n_outputs=output_dim,
n_reservoir=50,
sparsity=0.9,
spectral_radius=1,
noise=5,
input_scaling=True,
teacher_forcing=True,
random_state=42)
import matplotlib
matplotlib.use('Agg')
import pyESN
import matplotlib.pyplot as plt
import numpy as np
from numpy import genfromtxt
data = genfromtxt('lorenz_data_F8_new_std.csv', delimiter=",")
dataT = np.transpose(data)
np.shape(dataT)
esn = pyESN.ESN(n_inputs=1,
n_outputs=1,
n_reservoir=5000,
spectral_radius=1.5,
random_state=42)
trainN = 50000
testN = 2000
sol_matrix = np.zeros([testN, 40])
np.shape(sol_matrix)
for ptr in range(0, 40):
pred_training = esn.fit(np.ones(trainN), dataT[0:trainN, ptr])
prediction = esn.predict(np.ones(testN))
print("test error: \n" + str(
np.sqrt(
np.mean((prediction.flatten() -
dataT[trainN:trainN + testN, ptr])**2))))
def run(res, spr, reg):
esn = pyESN.ESN(n_inputs=1,
n_outputs=4,
n_reservoir=res,
spectral_radius=spr,
sparsity=0.002,
reg=reg,
random_state=42)
spin_off = 2000
trainN = 2200
testN = 2000
sol_matrix = np.zeros([testN, 64])
count = 0
for ptr in range(0, 16):
pred_training = esn.fit(np.ones(trainN - spin_off),
dataT[spin_off:trainN, count:count + 4])
prediction = esn.predict(np.ones(testN))
sol_matrix[:, count:count + 4] = (prediction)
count = count + 4
plt.figure(figsize=(15, 15))
for ptr in range(0, 32):
plt.subplot(8, 4, ptr + 1)
plt.plot(range(trainN - 1000, trainN + testN),
dataT[trainN - 1000:trainN + testN, ptr],
'k',
label="target system")
plt.plot(range(trainN, trainN + testN),
sol_matrix[:, ptr],
'y',
label="free running ESN")
lo, hi = plt.ylim()
plt.plot([trainN, trainN], [lo + np.spacing(1), hi - np.spacing(1)],
'k:')
plt.legend(loc=(0.61, 1.1), fontsize='x-small')
plt.ylabel('grid point ' + str(ptr), fontsize=14)
plt.savefig('time_series_' + str(res) + '_' + str(spr) + '_' + str(reg) +
'.png')
[t, grid] = np.meshgrid((np.arange(trainN, trainN + testN)),
np.arange(0, 64))
diff_mat = dataT[trainN:trainN + testN, :] - sol_matrix[0:testN, :]
v = np.linspace(-3, 3, 10, endpoint=True)
plt.figure(figsize=(45, 45))
plt.subplot(3, 1, 1)
v = np.linspace(-3, 3, 10, endpoint=True)
plt.contourf(np.transpose(t),
np.transpose(grid),
dataT[trainN:trainN + testN, :],
cmap=cm.jet,
vmin=-3,
vmax=3)
plt.colorbar(ticks=v)
plt.rcParams['xtick.labelsize'] = 40
plt.rcParams['ytick.labelsize'] = 40
plt.xlabel('$dt$', fontsize=40)
plt.ylabel('Grid Point', fontsize=40)
plt.title('Truth', fontsize=40)
plt.subplot(3, 1, 2)
plt.contourf(np.transpose(t),
np.transpose(grid),
sol_matrix[0:testN, :],
cmap=cm.jet,
vmin=-3,
vmax=3)
plt.colorbar(ticks=v)
#cbar.ax.set_ylabel('verbosity coefficient')
#plt.clim(-3,3)
plt.rcParams['xtick.labelsize'] = 40
plt.rcParams['ytick.labelsize'] = 40
plt.xlabel('$dt$', fontsize=40)
plt.ylabel('Grid Point', fontsize=40)
plt.title('Prediction', fontsize=40)
plt.subplot(3, 1, 3)
plt.contourf(np.transpose(t),
np.transpose(grid),
diff_mat,
cmap=cm.jet,
vmin=-3,
vmax=3)
plt.colorbar(ticks=v)
#cbar.ax.set_ylabel('verbosity coefficient')
plt.rcParams['xtick.labelsize'] = 40
plt.rcParams['ytick.labelsize'] = 40
plt.xlabel('$dt$', fontsize=40)
plt.ylabel('Grid Point', fontsize=40)
plt.title('Error Matrix', fontsize=40)
plt.savefig('contourplots_' + str(res) + '_' + str(spr) + '_' + str(reg) +
'.png')
data = genfromtxt('lorenz_data_F8_new_std.csv', delimiter=",")
dataT = np.transpose(data)
np.shape(dataT)
# trainN = 50000
trainN = 30000 # lover number for debugging on laptop
# testN = 2000
testN = 600 # lover number for debugging on laptop
sol_matrix = np.zeros([testN, 40])
np.shape(sol_matrix)
for ptr in range(0, 40):
esn = pyESN.ESN(
n_inputs=1,
n_outputs=1,
# n_reservoir=5000,
n_reservoir=160, # lover number for debugging on laptop
spectral_radius=0.9,
sparsity=0.002,
random_state=42)
pred_training = esn.fit(np.ones(trainN), dataT[0:trainN, ptr])
prediction = esn.predict(np.ones(testN))
print(
ptr, "test error: \n" + str(
np.sqrt(
np.mean(
(prediction.flatten() - dataT[trainN:trainN + testN, ptr])
**2))))
sol_matrix[:, ptr] = np.transpose(prediction)
np.shape(sol_matrix)
if len(sys.argv) != 2:
print("bad command line argumentas - pass the number of reservoirs")
exit(1)
reservoirs = int(sys.argv[1])
data = genfromtxt('lorenz_63_modified.csv', delimiter=",")
dataT = np.transpose(data)
dataT = np.transpose(dataT)
# dataT=dataT[2500:,:]
np.shape(dataT)
esn = pyESN.ESN(n_inputs=1,
n_outputs=3,
n_reservoir=reservoirs,
spectral_radius=0.9,
sparsity=0.002,
random_state=42)
spin_off = 0
trainN = 40000
testN = 2000
sol_matrix = np.zeros([testN, 3])
np.shape(sol_matrix)
pred_training = esn.fit(np.ones(trainN - spin_off),
dataT[spin_off:trainN, :],
inspect=True)
prediction = esn.predict(np.ones(testN))
import matplotlib
matplotlib.use('Agg')
import pyESN
import matplotlib.pyplot as plt
import numpy as np
from numpy import genfromtxt
from matplotlib import cm
data = genfromtxt('KS_data.csv', delimiter=",")
dataT = np.transpose(data)
esn = pyESN.ESN(n_inputs=1,
n_outputs=4,
n_reservoir=5000,
spectral_radius=0.95,
sparsity=0.002,
reg=0.0001,
random_state=42)
spin_off = 2000
trainN = 22000
testN = 2000
sol_matrix = np.zeros([testN, 64])
count = 0
for ptr in range(0, 16):
pred_training = esn.fit(np.ones(trainN - spin_off), dataT[spin_off:trainN,
count:count + 4])
prediction = esn.predict(np.ones(testN))
sol_matrix[:, count:count + 4] = (prediction)
