InvalidArgumentError: Graph execution error - python-3.x

Question

I am trying to apply LCM (linear coregionalization model), as a Gaussian process, to a CSV file dataset.
This dataset includes two inputs (FracYear, Auxiliar) and two outputs (VV,VH).
import gpflow as gpflow
import pandas as pd
import numpy as np
import xarray as xr
import matplotlib.pyplot as plt
plt.style.use('ggplot')
# %matplotlib inline
import seaborn as sns
np.random.seed(1)
def plot_gp(x, mu, var, color='k'):
plt.plot(x, mu, color=color, lw=2)
plt.plot(x, mu + 2*np.sqrt(var), '--', color=color)
plt.plot(x, mu - 2*np.sqrt(var), '--', color=color)
def plot(m):
xtest = np.linspace(0, 1, 300)[:,None]
line, = plt.plot(X1, Y1, 'x', mew=2)
mu, var = m.predict_f(np.hstack((xtest, np.zeros_like(xtest))))
plot_gp(xtest, mu, var, line.get_color())
line, = plt.plot(X2, Y2, 'x', mew=2)
mu, var = m.predict_f(np.hstack((xtest, np.ones_like(xtest))))
plot_gp(xtest, mu, var, line.get_color())
import pandas as pd
d = pd.read_csv('C://Users//Rick//Documents//UNI//PROJ//invento.csv',delimiter=',', header=None, skiprows=1, names=['FracYear', 'VH', 'VV'])
# Replace missing values with NaN
d.replace(-200.0, np.nan, inplace=True)
# Data preparation
# We start by generating some training data to fit the model with. For this example, we choose the following two correlated functions for our outputs:
# make a dataset with two outputs, correlated, heavy-tail noise. One has more noise than the other.
df = pd.DataFrame(data=d)
X1 = df['FracYear'] = pd.to_numeric(df['FracYear'])
X2 = df['Auxiliar'] = pd.to_numeric(df['Auxiliar'])
Y1 = df['VH'] = pd.to_numeric(df['VH'])
Y2 = df['VV'] = pd.to_numeric(df['VV'])
plt.plot(X1, Y1, 'x', mew=2)
plt.plot(X2, Y2, 'x', mew=2)
plt.show()
# Base Matern kernel
k1 = gpflow.kernels.Matern32(active_dims=[0])
# Build the coreg kernel
coreg = gpflow.kernels.Coregion(output_dim=2, rank=1, active_dims=[1])
kern = k1 * coreg
# Build Likelihood
lik = gpflow.likelihoods.SwitchedLikelihood([
gpflow.likelihoods.StudentT(), gpflow.likelihoods.StudentT()
])
# Augment the input with ones or zeros to indicate the required output dimension
X_augmented = np.vstack((np.hstack((X1, np.zeros_like(X1))), np.hstack((X2, np.ones_like(X2)))))
# Augment the Y data with ones or zeros that specify a likelihood from the list of likelihoods
Y_augmented = np.vstack((np.hstack((Y1, np.zeros_like(Y1))), np.hstack((Y2, np.ones_like(Y2)))))
# now buld the GP model as normal
m = gpflow.models.VGP((X_augmented, Y_augmented), kernel=kern, likelihood=lik)
# fit the covariance function parameters
#gpflow.train.ScipyOptimizer().minimize(m, maxiter=1000)
from gpflow.ci_utils import ci_niter
maxiter = ci_niter(10000)
gpflow.optimizers.Scipy().minimize(
m.training_loss, m.trainable_variables, options=dict(maxiter=maxiter), method="L-BFGS-B",
)
## Fit and plot
xtest = np.hstack([np.linspace(0, 1, 100)]*3)[:,None]
mu1, var1 = m.predict_f(np.hstack((xtest, np.zeros_like(xtest))))
mu2, var2 = m.predict_f(np.hstack((xtest, np.ones_like(xtest))))
plt.plot(X1, Y1, 'x', mew=2, color='r')
plt.plot(X2, Y2, 'x', mew=2, color='b')
plt.plot(np.linspace(0, 1, 100), np.reshape(mu2, [100,3]))
plt.plot(np.linspace(0, 1, 100), mu1, 'r')
plt.plot(np.linspace(0, 1, 100), mu2, 'b')
plt.show()
Please, note that I am using Colab. On the other hand, I am having issues when installing Tensorflow in Spyder. Having said this, I don't know which would be better: Colab, Spyder, Jupyter.
My doubt is the error resulting from the prompt with "gpflow.optimizers.Scipy().minimize".
The error is very long. It starts with the title of this thread, and ends like this: "Node: 'GatherV2_2'
indices[0] = 2019 is not in [0, 2)
[[{{node GatherV2_2}}]] [Op:__inference__tf_eval_9447]"
If anyone has any idea about this, please notice me. Also, if any of you has a LCM code which works for csv files, it may be interesting for me to keep it an eye.
Thanks!!