Last active
October 29, 2020 18:01
-
-
Save gurbain/ba52af78d7be6eb2a23f48af15da2ce0 to your computer and use it in GitHub Desktop.
pyESN + keras readout divergence
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# Scipy libraries | |
import numpy as np | |
import os | |
from matplotlib import pyplot as plt | |
plt.style.use('fivethirtyeight') | |
plt.rc('lines', linewidth= 1) | |
plt.rc('text', usetex=False) | |
plt.rc('axes', facecolor='white') | |
plt.rc('savefig', facecolor='white') | |
plt.rc('figure', autolayout=True) | |
# pyESN libraries | |
from pyESN import ESN | |
# Keras libraries | |
from keras.optimizers import Adam | |
from keras.models import Sequential | |
from keras.layers import Dense | |
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' | |
# Load the data | |
data = np.load('mackey_glass_t17.npy') | |
# Declare main model and data parameters | |
n_res = 300 | |
n_out = 1 | |
n_in = 1 | |
trainlen = 2000 | |
future = 2000 | |
# Create the keras network for readout layer | |
nn = Sequential() | |
nn.add(Dense(100, input_dim=n_res+n_in, activation='relu')) | |
nn.add(Dense(1, activation='linear')) | |
nn.compile(loss='mse', optimizer='adam', metrics=['mse']) | |
nn.summary() | |
# Create the reservoir | |
esn = ESN(n_inputs = n_in, | |
n_outputs = n_out, | |
n_reservoir = n_res, | |
noise = 0.01, | |
spectral_radius = 1.4, | |
teacher_forcing = True, | |
random_state = 42, | |
keras_model = nn) | |
# Train | |
pred_training = esn.fit(np.ones(trainlen),data[:trainlen], | |
epochs=20, verbose=2) | |
print("Train error: \n" + \ | |
str(np.sqrt(np.mean((pred_training.flatten() - data[:trainlen])**2)))) | |
# Test | |
prediction = esn.predict(np.ones(future)) | |
print("Test error: \n" + \ | |
str(np.sqrt(np.mean((prediction.flatten() - data[trainlen:trainlen +future])**2)))) | |
# Plot results | |
plt.figure(figsize=(11,1.5)) | |
plt.plot(range(0,trainlen),data[0:trainlen], | |
lineStyle="--", linewidth=2, label="Train Prediction") | |
plt.plot(range(0,trainlen+future),data[0:trainlen+future], | |
label="Target System") | |
plt.plot(range(trainlen,trainlen+future),prediction, | |
label="Test Prediction") | |
lo,hi = plt.ylim() | |
plt.plot([trainlen,trainlen],[lo+np.spacing(1),hi-np.spacing(1)],'k:') | |
plt.legend(fontsize='x-small') | |
plt.show() |
Hi Gabriel,
I removed the comment because I have found it and I did not want to annoy you,
but, thank you!
It is here in case you need it in the future: cknd/pyESN@55cb273
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi Francesca,
I honnestly can't remember! :/ I completely forgot about this gist. I dates from August 2018, is it not then the one form here: https://github.com/cknd/pyESN? I think it was last updated slightly before I wrote this piece of code.