I would like to perform some multivariant regression using gaussian process regression as implemented in GPflow using version 2.
Installed with pip install gpflow==2.0.0rc1
Below is some example code that generates some 2D data and then attempts to fit it with using GPR and the finally computes the difference
between the true input data and the GPR prediction.
Eventually I would like to extend to higher dimensions
and do tests against a validation set to check for over-fitting
and experiment with other kernels and "Automatic Relevance Determination"
but understanding how to get this to work is the first step.
Thanks!
The following code snippet will work in a jupyter notebook.
import gpflow
import numpy as np
import matplotlib
from gpflow.utilities import print_summary
%matplotlib inline
matplotlib.rcParams['figure.figsize'] = (12, 6)
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
def gen_data(X, Y):
"""
make some fake data.
X, Y are np.ndarrays with shape (N,) where
N is the number of samples.
"""
ys = []
for x0, x1 in zip(X,Y):
y = x0 * np.sin(x0*10)
y = x1 * np.sin(x0*10)
y += 1
ys.append(y)
return np.array(ys)
# generate some fake data
x = np.linspace(0, 1, 20)
X, Y = np.meshgrid(x, x)
X = X.ravel()
Y = Y.ravel()
z = gen_data(X, Y)
#note X.shape, Y.shape and z.shape
#are all (400,) for this case.
# if you would like to plot the data you can do the following
fig = plt.figure()
ax = Axes3D(fig)
ax.scatter(X, Y, z, s=100, c='k')
# had to set this
# to avoid the following error
# tensorflow.python.framework.errors_impl.InvalidArgumentError: Cholesky decomposition was not successful. The input might not be valid. [Op:Cholesky]
gpflow.config.set_default_positive_minimum(1e-7)
# setup the kernel
k = gpflow.kernels.Matern52()
# set up GPR model
# I think the shape of the independent data
# should be (400, 2) for this case
XY = np.column_stack([[X, Y]]).T
print(XY.shape) # this will be (400, 2)
m = gpflow.models.GPR(data=(XY, z), kernel=k, mean_function=None)
# optimise hyper-parameters
opt = gpflow.optimizers.Scipy()
def objective_closure():
return - m.log_marginal_likelihood()
opt_logs = opt.minimize(objective_closure,
m.trainable_variables,
options=dict(maxiter=100)
)
# predict training set
mean, var = m.predict_f(XY)
print(mean.numpy().shape)
# (400, 400)
# I would expect this to be (400,)
# If it was then I could compute the difference
# between the true data and the GPR prediction
# `diff = mean - z`
# but because the shape is not as expected this of course
# won't work.
The shape of z must be (N, 1), whereas in your case it is (N,). However, this is a missing check in GPflow and not your fault.
Related
I have found few radial basis functions like BasisExpansionFunction, Surrogates.jl, ScatteredInterpolation in Julia.
However, I am unable to replicate the results from python's scipy.interpolate.rbf() function.
Python Example
from scipy.interpolate import Rbf
import numpy as np
xs = np.arange(10)
ys = xs**2 + np.sin(xs) + 1
interp_func = Rbf(xs, ys) # By default RbF uses Multiquadratic function
newarr = interp_func(np.arange(2.1, 3, 0.1))
print(newarr)
What is correct approach to replicate the above example in Julia?
The first tutorial in Surrogates.jl shows how to make and interpolate a radial basis function.
using Surrogates
using LinearAlgebra
f = x -> x[1]*x[2]
lb = [1.0,2.0]
ub = [10.0,8.5]
x = sample(50,lb,ub,SobolSample())
y = f.(x)
my_radial_basis = RadialBasis(x,y,lb,ub)
#I want an approximation at (1.0,1.4)
approx = my_radial_basis((1.0,1.4))
In GPflow I have multiple time series and the sampling times are not aligned across time series, and the time series may have different length (longitudinal data). I assume that they are independent realizations from the same GP. What is the right way to handle this with svgp, and more generally with GPflow? Do i need to use coregionalization? The coregionalization notebook assumed correlated trajectories, while I want shared mean/kernel but independent.
Yes, the Coregion kernel implemented in GPflow is what you can use for your problem.
Let's set up some data from the generative model you describe, with different lengths for the timeseries:
import numpy as np
import gpflow
import matplotlib.pyplot as plt
Ns = [80, 90, 100] # number of observations for three different realizations
Xs = [np.random.uniform(0, 10, size=N) for N in Ns] # observation locations
# three different draws from the same GP:
k = gpflow.kernels.Matern52(variance=2.0, lengthscales=0.5) # kernel
Ks = [k(X[:, None]) for X in Xs]
Ls = [np.linalg.cholesky(K) for K in Ks]
vs = [np.random.randn(N, 1) for N in Ns]
fs = [(L # v).squeeze(axis=-1) for L, v in zip(Ls, vs)]
To actually set up the training data for the gpflow GP model:
# output indicator for the observations: which timeseries is this?
os = [o * np.ones(N) for o, N in enumerate(Ns)] # [0 ... 0, 1 ... 1, 2 ... 2]
# now assemble the three timeseries in single data set:
allX = np.concatenate(Xs)
allo = np.concatenate(os)
allf = np.concatenate(fs)
X = np.c_[allX, allo]
Y = allf[:, None]
assert X.shape == (sum(Ns), 2)
assert Y.shape == (sum(Ns), 1)
# now let's set up a copy of the original kernel:
k2 = gpflow.kernels.Matern52(active_dims=[0]) # the same as k above, but with different hyperparameters
# and a Coregionalization kernel that effectively says they are all independent:
kc = gpflow.kernels.Coregion(output_dim=len(Ns), rank=1, active_dims=[1])
kc.W.assign(np.zeros(kc.W.shape))
kc.kappa.assign(np.ones(kc.kappa.shape))
gpflow.set_trainable(kc, False) # we want W and kappa fixed
The Coregion kernel defines a covariance matrix B = W Wᵀ + diag(kappa), so by setting W=0 we prescribe zero correlations (independent realizations) and kappa=1 (actually the default) ensures that the variance hyperparameter of the copy of the original kernel remains interpretable.
Now construct the actual model and optimize hyperparameters:
k2c = k2 * kc
m = gpflow.models.GPR((X, Y), k2c, noise_variance=1e-5)
opt = gpflow.optimizers.Scipy()
opt.minimize(m.training_loss, m.trainable_variables, compile=False)
which recovers the initial variance and lengthscale hyperparameters pretty well.
If you want to predict, you have to provide the extra "output" column in the Xnew argument to m.predict_f(), e.g. as follows:
Xtest = np.linspace(0, 10, 100)
Xtest_augmented = np.c_[Xtest, np.zeros_like(Xtest)]
f_mean, f_var = m.predict_f(Xtest_augmented)
(whether you set the output column to 0, 1, or 2 does not matter, as we set them all to be the same with our choice of W and kappa).
If your input was more than one-dimensional, you could set
active_dims=list(range(X.shape[1] - 1)) for the first kernel(s) and active_dims=[X.shape[1]-1] for the Coregion kernel.
I have a set of about 33K (x,y,z) points in a csv file and would like to convert this to a grid of density values using scipy.stats.gaussian_kde. I have not been able to find a way to convert this point cloud array into an appropriate input format for the gaussian_kde function (and then take the output of this and convert it into a density value grid). Can anyone provide sample code?
Here's an example with some comments which may be of use. gaussian_kde wants the data and points to be row stacked, ie. (# ndim, # num values), as per the docs. In your case you would row_stack([x, y, z]) such that the shape is (3, 33000).
from scipy.stats import gaussian_kde
import numpy as np
import matplotlib.pyplot as plt
# simulate some data
n = 33000
x = np.random.randn(n)
y = np.random.randn(n) * 2
# data must be stacked as (# ndim, # n values) as per docs.
data = np.row_stack((x, y))
# perform KDE
kernel = gaussian_kde(data)
# create grid over which to evaluate KDE
s = np.linspace(-8, 8, 128)
grid = np.meshgrid(s, s)
# again KDE needs points to be row_stacked
grid_points = np.row_stack([g.ravel() for g in grid])
# evaluate KDE and reshape result correctly
Z = kernel(grid_points)
Z = Z.reshape(grid[0].shape)
# plot KDE as image and overlay some data points
fig, ax = plt.subplots()
ax.matshow(Z, extent=(s.min(), s.max(), s.min(), s.max()))
ax.plot(x[::10], y[::10], 'w.', ms=1, alpha=0.3)
ax.set_xlim(s.min(), s.max())
ax.set_ylim(s.min(), s.max())
After reading about how to solve an ODE with neural networks following the paper Neural Ordinary Differential Equations and the blog that uses the library JAX I tried to do the same thing with "plain" Pytorch but found a point rather "obscure": How to properly use the partial derivative of a function (in this case the model) w.r.t one of the input parameters.
To resume the problem at hand as shown in 2 it is intended to solve the ODE y' = -2*x*y with the condition y(x=0) = 1 in the domain -2 <= x <= 2. Instead of using finite differences the solution is replaced by a NN as y(x) = NN(x) with a single layer with 10 nodes.
I managed to (more or less) replicate the blog with the following code
import torch
import torch.nn as nn
from torch import optim
import matplotlib.pyplot as plt
import numpy as np
# Define the NN model to solve the problem
class Model(nn.Module):
def __init__(self):
super(Model, self).__init__()
self.lin1 = nn.Linear(1,10)
self.lin2 = nn.Linear(10,1)
def forward(self, x):
x = torch.sigmoid(self.lin1(x))
x = torch.sigmoid(self.lin2(x))
return x
model = Model()
# Define loss_function from the Ordinary differential equation to solve
def ODE(x,y):
dydx, = torch.autograd.grad(y, x,
grad_outputs=y.data.new(y.shape).fill_(1),
create_graph=True, retain_graph=True)
eq = dydx + 2.* x * y # y' = - 2x*y
ic = model(torch.tensor([0.])) - 1. # y(x=0) = 1
return torch.mean(eq**2) + ic**2
loss_func = ODE
# Define the optimization
# opt = optim.SGD(model.parameters(), lr=0.1, momentum=0.99,nesterov=True) # Equivalent to blog
opt = optim.Adam(model.parameters(),lr=0.1,amsgrad=True) # Got faster convergence with Adam using amsgrad
# Define reference grid
x_data = torch.linspace(-2.0,2.0,401,requires_grad=True)
x_data = x_data.view(401,1) # reshaping the tensor
# Iterative learning
epochs = 1000
for epoch in range(epochs):
opt.zero_grad()
y_trial = model(x_data)
loss = loss_func(x_data, y_trial)
loss.backward()
opt.step()
if epoch % 100 == 0:
print('epoch {}, loss {}'.format(epoch, loss.item()))
# Plot Results
plt.plot(x_data.data.numpy(), np.exp(-x_data.data.numpy()**2), label='exact')
plt.plot(x_data.data.numpy(), y_data.data.numpy(), label='approx')
plt.legend()
plt.show()
From here I manage to get the results as shown in the fig.
enter image description here
The problems is that at the definition of the ODE functional, instead of passing (x,y) I would rather prefer to pass something like (x,fun) (where fun is my model) such that the partial derivative and specific evaluations of the model can be done with a call . So, something like
def ODE(x,fun):
dydx, = "grad of fun w.r.t x as a function"
eq = dydx(x) + 2.* x * fun(x) # y' = - 2x*y
ic = fun( torch.tensor([0.]) ) - 1. # y(x=0) = 1
return torch.mean(eq**2) + ic**2
Any ideas? Thanks in advance
EDIT:
After some trials I found a way to pass the model as an input but found another strange behavior... The new problem is to solve the ODE y'' = -2 with the BC y(x=-2) = -1 and y(x=2) = 1, for which the analytical solution is y(x) = -x^2+x/2+4
Let's modify a bit the previous code as:
import torch
import torch.nn as nn
from torch import optim
import matplotlib.pyplot as plt
import numpy as np
# Define the NN model to solve the equation
class Model(nn.Module):
def __init__(self):
super(Model, self).__init__()
self.lin1 = nn.Linear(1,10)
self.lin2 = nn.Linear(10,1)
def forward(self, x):
y = torch.sigmoid(self.lin1(x))
z = torch.sigmoid(self.lin2(y))
return z
model = Model()
# Define loss_function from the Ordinary differential equation to solve
def ODE(x,fun):
y = fun(x)
dydx = torch.autograd.grad(y, x,
grad_outputs=y.data.new(y.shape).fill_(1),
create_graph=True, retain_graph=True)[0]
d2ydx2 = torch.autograd.grad(dydx, x,
grad_outputs=dydx.data.new(dydx.shape).fill_(1),
create_graph=True, retain_graph=True)[0]
eq = d2ydx2 + torch.tensor([ 2.]) # y'' = - 2
bc1 = fun(torch.tensor([-2.])) - torch.tensor([-1.]) # y(x=-2) = -1
bc2 = fun(torch.tensor([ 2.])) - torch.tensor([ 1.]) # y(x= 2) = 1
return torch.mean(eq**2) + bc1**2 + bc2**2
loss_func = ODE
So, here I passed the model as argument and managed to derive twice... so far so good. BUT, using the sigmoid function for this case is not only not necessary but also gives a result that is far from the analytical one.
If I change the NN for:
class Model(nn.Module):
def __init__(self):
super(Model, self).__init__()
self.lin1 = nn.Linear(1,1)
self.lin2 = nn.Linear(1,1)
def forward(self, x):
y = self.lin1(x)
z = self.lin2(y)
return z
In which case I would expect to optimize a double pass through two linear functions that would retrieve a 2nd order function ... I get the error:
RuntimeError: One of the differentiated Tensors appears to not have been used in the graph. Set allow_unused=True if this is the desired behavior.
Adding the option to the definition of dydx doesn't solve the problem, and adding it to d2ydx2 gives a NoneType definition.
Is there something wrong with the layers as they are?
Quick Solution:
add allow_unused=True to .grad functions. So, change
dydx = torch.autograd.grad(
y, x,
grad_outputs=y.data.new(y.shape).fill_(1),
create_graph=True, retain_graph=True)[0]
d2ydx2 = torch.autograd.grad(dydx, x, grad_outputs=dydx.data.new(
dydx.shape).fill_(1), create_graph=True, retain_graph=True)[0]
To
dydx = torch.autograd.grad(
y, x,
grad_outputs=y.data.new(y.shape).fill_(1),
create_graph=True, retain_graph=True, allow_unused=True)[0]
d2ydx2 = torch.autograd.grad(dydx, x, grad_outputs=dydx.data.new(
dydx.shape).fill_(1), create_graph=True, retain_graph=True, allow_unused=True)[0]
More explanation:
See what allow_unused do:
allow_unused (bool, optional): If ``False``, specifying inputs that were not
used when computing outputs (and therefore their grad is always zero)
is an error. Defaults to ``False``.
So, if you try to differentiate w.r.t to a variable that is not in being used to compute the value, it will give an error. Also, note that error only occurs when you use linear layers.
This is because when you use linear layers, you have y=W1*W2*x + b = Wx+b and dy/dx is not a function of x, it is simply W. So when you try to differentiate dy/dx w.r.t x it throws an error. This error goes away as soon as you use sigmoid because then dy/dx will be a function of x. To avoid the error, either make sure dy/dx is a function of x or use allow_unused=True
I'm trying to convolve two 1D tensors in Keras.
I get two inputs from other models:
x - of length 100
ker - of length 5
I would like to get the 1D convolution of x using the kernel ker.
I wrote a Lambda layer to do it:
import tensorflow as tf
def convolve1d(x):
y = tf.nn.conv1d(value=x[0], filters=x[1], padding='VALID', stride=1)
return y
x = Input(shape=(100,))
ker = Input(shape=(5,))
y = Lambda(convolve1d)([x,ker])
model = Model([x,ker], [y])
I get the following error:
ValueError: Shape must be rank 4 but is rank 3 for 'lambda_67/conv1d/Conv2D' (op: 'Conv2D') with input shapes: [?,1,100], [1,?,5].
Can anyone help me understand how to fix it?
It was much harder than I expected because Keras and Tensorflow don't expect any batch dimension in the convolution kernel so I had to write the loop over the batch dimension myself, which requires to specify batch_shape instead of just shape in the Input layer. Here it is :
import numpy as np
import tensorflow as tf
import keras
from keras import backend as K
from keras import Input, Model
from keras.layers import Lambda
def convolve1d(x):
input, kernel = x
output_list = []
if K.image_data_format() == 'channels_last':
kernel = K.expand_dims(kernel, axis=-2)
else:
kernel = K.expand_dims(kernel, axis=0)
for i in range(batch_size): # Loop over batch dimension
output_temp = tf.nn.conv1d(value=input[i:i+1, :, :],
filters=kernel[i, :, :],
padding='VALID',
stride=1)
output_list.append(output_temp)
print(K.int_shape(output_temp))
return K.concatenate(output_list, axis=0)
batch_input_shape = (1, 100, 1)
batch_kernel_shape = (1, 5, 1)
x = Input(batch_shape=batch_input_shape)
ker = Input(batch_shape=batch_kernel_shape)
y = Lambda(convolve1d)([x,ker])
model = Model([x, ker], [y])
a = np.ones(batch_input_shape)
b = np.ones(batch_kernel_shape)
c = model.predict([a, b])
In the current state :
It doesn't work for inputs (x) with multiple channels.
If you provide several filters, you get as many outputs, each being the convolution of the input with the corresponding kernel.
From given code it is difficult to point out what you mean when you say
is it possible
But if what you mean is to merge two layers and feed merged layer to convulation, yes it is possible.
x = Input(shape=(100,))
ker = Input(shape=(5,))
merged = keras.layers.concatenate([x,ker], axis=-1)
y = K.conv1d(merged, 'same')
model = Model([x,ker], y)
EDIT:
#user2179331 thanks for clarifying your intention. Now you are using Lambda Class incorrectly, that is why the error message is showing.
But what you are trying to do can be achieved using keras.backend layers.
Though be noted that when using lower level layers you will lose some higher level abstraction. E.g when using keras.backend.conv1d you need to have input shape of (BATCH_SIZE,width, channels) and kernel with shape of (kernel_size,input_channels,output_channels). So in your case let as assume the x has channels of 1(input channels ==1) and y also have the same number of channels(output channels == 1).
So your code now can be refactored as follows
from keras import backend as K
def convolve1d(x,kernel):
y = K.conv1d(x,kernel, padding='valid', strides=1,data_format="channels_last")
return y
input_channels = 1
output_channels = 1
kernel_width = 5
input_width = 100
ker = K.variable(K.random_uniform([kernel_width,input_channels,output_channels]),K.floatx())
x = Input(shape=(input_width,input_channels)
y = convolve1d(x,ker)
I guess I have understood what you mean. Given the wrong example code below:
input_signal = Input(shape=(L), name='input_signal')
input_h = Input(shape=(N), name='input_h')
faded= Lambda(lambda x: tf.nn.conv1d(input, x))(input_h)
You want to convolute each signal vector with different fading coefficients vector.
The 'conv' operation in TensorFlow, etc. tf.nn.conv1d, only support a fixed value kernel. Therefore, the code above can not run as you want.
I have no idea, too. The code you given can run normally, however, it is too complex and not efficient. In my idea, another feasible but also inefficient way is to multiply with the Toeplitz matrix whose row vector is the shifted fading coefficients vector. When the signal vector is too long, the matrix will be extremely large.