Commit 91f480e9 authored by Martin Reinecke's avatar Martin Reinecke
Browse files

5->6

parent 14052dd3
...@@ -37,7 +37,7 @@ build_docker_from_cache: ...@@ -37,7 +37,7 @@ build_docker_from_cache:
test_serial: test_serial:
stage: test stage: test
script: script:
- pytest-3 -q --cov=nifty5 test - pytest-3 -q --cov=nifty6 test
- > - >
python3 -m coverage report --omit "*plot*,*distributed_do*" | tee coverage.txt python3 -m coverage report --omit "*plot*,*distributed_do*" | tee coverage.txt
- > - >
...@@ -59,7 +59,7 @@ pages: ...@@ -59,7 +59,7 @@ pages:
paths: paths:
- public - public
only: only:
- NIFTy_5 - NIFTy_6
before_script: before_script:
......
NIFTy - Numerical Information Field Theory NIFTy - Numerical Information Field Theory
========================================== ==========================================
[![build status](https://gitlab.mpcdf.mpg.de/ift/NIFTy/badges/NIFTy_5/build.svg)](https://gitlab.mpcdf.mpg.de/ift/NIFTy/commits/NIFTy_5) [![build status](https://gitlab.mpcdf.mpg.de/ift/NIFTy/badges/NIFTy_6/build.svg)](https://gitlab.mpcdf.mpg.de/ift/NIFTy/commits/NIFTy_6)
[![coverage report](https://gitlab.mpcdf.mpg.de/ift/NIFTy/badges/NIFTy_5/coverage.svg)](https://gitlab.mpcdf.mpg.de/ift/NIFTy/commits/NIFTy_5) [![coverage report](https://gitlab.mpcdf.mpg.de/ift/NIFTy/badges/NIFTy_6/coverage.svg)](https://gitlab.mpcdf.mpg.de/ift/NIFTy/commits/NIFTy_6)
**NIFTy** project homepage: **NIFTy** project homepage:
[http://ift.pages.mpcdf.de/nifty](http://ift.pages.mpcdf.de/nifty) [http://ift.pages.mpcdf.de/nifty](http://ift.pages.mpcdf.de/nifty)
...@@ -59,8 +59,8 @@ Optional dependencies: ...@@ -59,8 +59,8 @@ Optional dependencies:
### Sources ### Sources
The current version of Nifty5 can be obtained by cloning the repository and The current version of NIFTy6 can be obtained by cloning the repository and
switching to the NIFTy_5 branch: switching to the NIFTy_6 branch:
git clone https://gitlab.mpcdf.mpg.de/ift/nifty.git git clone https://gitlab.mpcdf.mpg.de/ift/nifty.git
...@@ -69,10 +69,10 @@ switching to the NIFTy_5 branch: ...@@ -69,10 +69,10 @@ switching to the NIFTy_5 branch:
In the following, we assume a Debian-based distribution. For other In the following, we assume a Debian-based distribution. For other
distributions, the "apt" lines will need slight changes. distributions, the "apt" lines will need slight changes.
NIFTy5 and its mandatory dependencies can be installed via: NIFTy6 and its mandatory dependencies can be installed via:
sudo apt-get install git python3 python3-pip python3-dev sudo apt-get install git python3 python3-pip python3-dev
pip3 install --user git+https://gitlab.mpcdf.mpg.de/ift/nifty.git@NIFTy_5 pip3 install --user git+https://gitlab.mpcdf.mpg.de/ift/nifty.git@NIFTy_6
pip3 install --user git+https://gitlab.mpcdf.mpg.de/mtr/pypocketfft pip3 install --user git+https://gitlab.mpcdf.mpg.de/mtr/pypocketfft
Plotting support is added via: Plotting support is added via:
...@@ -100,7 +100,7 @@ To run the tests, additional packages are required: ...@@ -100,7 +100,7 @@ To run the tests, additional packages are required:
Afterwards the tests (including a coverage report) can be run using the Afterwards the tests (including a coverage report) can be run using the
following command in the repository root: following command in the repository root:
pytest-3 --cov=nifty5 test pytest-3 --cov=nifty6 test
### First Steps ### First Steps
......
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
# A NIFTy demonstration # A NIFTy demonstration
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
## IFT: Big Picture ## IFT: Big Picture
IFT starting point: IFT starting point:
$$d = Rs+n$$ $$d = Rs+n$$
Typically, $s$ is a continuous field, $d$ a discrete data vector. Particularly, $R$ is not invertible. Typically, $s$ is a continuous field, $d$ a discrete data vector. Particularly, $R$ is not invertible.
IFT aims at **inverting** the above uninvertible problem in the **best possible way** using Bayesian statistics. IFT aims at **inverting** the above uninvertible problem in the **best possible way** using Bayesian statistics.
## NIFTy ## NIFTy
NIFTy (Numerical Information Field Theory) is a Python framework in which IFT problems can be tackled easily. NIFTy (Numerical Information Field Theory) is a Python framework in which IFT problems can be tackled easily.
Main Interfaces: Main Interfaces:
- **Spaces**: Cartesian, 2-Spheres (Healpix, Gauss-Legendre) and their respective harmonic spaces. - **Spaces**: Cartesian, 2-Spheres (Healpix, Gauss-Legendre) and their respective harmonic spaces.
- **Fields**: Defined on spaces. - **Fields**: Defined on spaces.
- **Operators**: Acting on fields. - **Operators**: Acting on fields.
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
## Wiener Filter: Formulae ## Wiener Filter: Formulae
### Assumptions ### Assumptions
- $d=Rs+n$, $R$ linear operator. - $d=Rs+n$, $R$ linear operator.
- $\mathcal P (s) = \mathcal G (s,S)$, $\mathcal P (n) = \mathcal G (n,N)$ where $S, N$ are positive definite matrices. - $\mathcal P (s) = \mathcal G (s,S)$, $\mathcal P (n) = \mathcal G (n,N)$ where $S, N$ are positive definite matrices.
### Posterior ### Posterior
The Posterior is given by: The Posterior is given by:
$$\mathcal P (s|d) \propto P(s,d) = \mathcal G(d-Rs,N) \,\mathcal G(s,S) \propto \mathcal G (s-m,D) $$ $$\mathcal P (s|d) \propto P(s,d) = \mathcal G(d-Rs,N) \,\mathcal G(s,S) \propto \mathcal G (s-m,D) $$
where where
$$\begin{align} $$\begin{align}
m &= Dj \\ m &= Dj \\
D^{-1}&= (S^{-1} +R^\dagger N^{-1} R )\\ D^{-1}&= (S^{-1} +R^\dagger N^{-1} R )\\
j &= R^\dagger N^{-1} d j &= R^\dagger N^{-1} d
\end{align}$$ \end{align}$$
Let us implement this in NIFTy! Let us implement this in NIFTy!
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
## Wiener Filter: Example ## Wiener Filter: Example
- We assume statistical homogeneity and isotropy. Therefore the signal covariance $S$ is diagonal in harmonic space, and is described by a one-dimensional power spectrum, assumed here as $$P(k) = P_0\,\left(1+\left(\frac{k}{k_0}\right)^2\right)^{-\gamma /2},$$ - We assume statistical homogeneity and isotropy. Therefore the signal covariance $S$ is diagonal in harmonic space, and is described by a one-dimensional power spectrum, assumed here as $$P(k) = P_0\,\left(1+\left(\frac{k}{k_0}\right)^2\right)^{-\gamma /2},$$
with $P_0 = 0.2, k_0 = 5, \gamma = 4$. with $P_0 = 0.2, k_0 = 5, \gamma = 4$.
- $N = 0.2 \cdot \mathbb{1}$. - $N = 0.2 \cdot \mathbb{1}$.
- Number of data points $N_{pix} = 512$. - Number of data points $N_{pix} = 512$.
- reconstruction in harmonic space. - reconstruction in harmonic space.
- Response operator: - Response operator:
$$R = FFT_{\text{harmonic} \rightarrow \text{position}}$$ $$R = FFT_{\text{harmonic} \rightarrow \text{position}}$$
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
N_pixels = 512 # Number of pixels N_pixels = 512 # Number of pixels
def pow_spec(k): def pow_spec(k):
P0, k0, gamma = [.2, 5, 4] P0, k0, gamma = [.2, 5, 4]
return P0 / ((1. + (k/k0)**2)**(gamma / 2)) return P0 / ((1. + (k/k0)**2)**(gamma / 2))
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
## Wiener Filter: Implementation ## Wiener Filter: Implementation
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Import Modules ### Import Modules
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
import numpy as np import numpy as np
np.random.seed(40) np.random.seed(40)
import nifty5 as ift import nifty6 as ift
import matplotlib.pyplot as plt import matplotlib.pyplot as plt
%matplotlib inline %matplotlib inline
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Implement Propagator ### Implement Propagator
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
def Curvature(R, N, Sh): def Curvature(R, N, Sh):
IC = ift.GradientNormController(iteration_limit=50000, IC = ift.GradientNormController(iteration_limit=50000,
tol_abs_gradnorm=0.1) tol_abs_gradnorm=0.1)
# WienerFilterCurvature is (R.adjoint*N.inverse*R + Sh.inverse) plus some handy # WienerFilterCurvature is (R.adjoint*N.inverse*R + Sh.inverse) plus some handy
# helper methods. # helper methods.
return ift.WienerFilterCurvature(R,N,Sh,iteration_controller=IC,iteration_controller_sampling=IC) return ift.WienerFilterCurvature(R,N,Sh,iteration_controller=IC,iteration_controller_sampling=IC)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Conjugate Gradient Preconditioning ### Conjugate Gradient Preconditioning
- $D$ is defined via: - $D$ is defined via:
$$D^{-1} = \mathcal S_h^{-1} + R^\dagger N^{-1} R.$$ $$D^{-1} = \mathcal S_h^{-1} + R^\dagger N^{-1} R.$$
In the end, we want to apply $D$ to $j$, i.e. we need the inverse action of $D^{-1}$. This is done numerically (algorithm: *Conjugate Gradient*). In the end, we want to apply $D$ to $j$, i.e. we need the inverse action of $D^{-1}$. This is done numerically (algorithm: *Conjugate Gradient*).
<!-- <!--
- One can define the *condition number* of a non-singular and normal matrix $A$: - One can define the *condition number* of a non-singular and normal matrix $A$:
$$\kappa (A) := \frac{|\lambda_{\text{max}}|}{|\lambda_{\text{min}}|},$$ $$\kappa (A) := \frac{|\lambda_{\text{max}}|}{|\lambda_{\text{min}}|},$$
where $\lambda_{\text{max}}$ and $\lambda_{\text{min}}$ are the largest and smallest eigenvalue of $A$, respectively. where $\lambda_{\text{max}}$ and $\lambda_{\text{min}}$ are the largest and smallest eigenvalue of $A$, respectively.
- The larger $\kappa$ the slower Conjugate Gradient. - The larger $\kappa$ the slower Conjugate Gradient.
- By default, conjugate gradient solves: $D^{-1} m = j$ for $m$, where $D^{-1}$ can be badly conditioned. If one knows a non-singular matrix $T$ for which $TD^{-1}$ is better conditioned, one can solve the equivalent problem: - By default, conjugate gradient solves: $D^{-1} m = j$ for $m$, where $D^{-1}$ can be badly conditioned. If one knows a non-singular matrix $T$ for which $TD^{-1}$ is better conditioned, one can solve the equivalent problem:
$$\tilde A m = \tilde j,$$ $$\tilde A m = \tilde j,$$
where $\tilde A = T D^{-1}$ and $\tilde j = Tj$. where $\tilde A = T D^{-1}$ and $\tilde j = Tj$.
- In our case $S^{-1}$ is responsible for the bad conditioning of $D$ depending on the chosen power spectrum. Thus, we choose - In our case $S^{-1}$ is responsible for the bad conditioning of $D$ depending on the chosen power spectrum. Thus, we choose
$$T = \mathcal F^\dagger S_h^{-1} \mathcal F.$$ $$T = \mathcal F^\dagger S_h^{-1} \mathcal F.$$
--> -->
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Generate Mock data ### Generate Mock data
- Generate a field $s$ and $n$ with given covariances. - Generate a field $s$ and $n$ with given covariances.
- Calculate $d$. - Calculate $d$.
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
s_space = ift.RGSpace(N_pixels) s_space = ift.RGSpace(N_pixels)
h_space = s_space.get_default_codomain() h_space = s_space.get_default_codomain()
HT = ift.HarmonicTransformOperator(h_space, target=s_space) HT = ift.HarmonicTransformOperator(h_space, target=s_space)
# Operators # Operators
Sh = ift.create_power_operator(h_space, power_spectrum=pow_spec) Sh = ift.create_power_operator(h_space, power_spectrum=pow_spec)
R = HT #*ift.create_harmonic_smoothing_operator((h_space,), 0, 0.02) R = HT #*ift.create_harmonic_smoothing_operator((h_space,), 0, 0.02)
# Fields and data # Fields and data
sh = Sh.draw_sample() sh = Sh.draw_sample()
noiseless_data=R(sh) noiseless_data=R(sh)
noise_amplitude = np.sqrt(0.2) noise_amplitude = np.sqrt(0.2)
N = ift.ScalingOperator(noise_amplitude**2, s_space) N = ift.ScalingOperator(noise_amplitude**2, s_space)
n = ift.Field.from_random(domain=s_space, random_type='normal', n = ift.Field.from_random(domain=s_space, random_type='normal',
std=noise_amplitude, mean=0) std=noise_amplitude, mean=0)
d = noiseless_data + n d = noiseless_data + n
j = R.adjoint_times(N.inverse_times(d)) j = R.adjoint_times(N.inverse_times(d))
curv = Curvature(R=R, N=N, Sh=Sh) curv = Curvature(R=R, N=N, Sh=Sh)
D = curv.inverse D = curv.inverse
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Run Wiener Filter ### Run Wiener Filter
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
m = D(j) m = D(j)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Signal Reconstruction ### Signal Reconstruction
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
# Get signal data and reconstruction data # Get signal data and reconstruction data
s_data = HT(sh).to_global_data() s_data = HT(sh).to_global_data()
m_data = HT(m).to_global_data() m_data = HT(m).to_global_data()
d_data = d.to_global_data() d_data = d.to_global_data()
plt.figure(figsize=(15,10)) plt.figure(figsize=(15,10))
plt.plot(s_data, 'r', label="Signal", linewidth=3) plt.plot(s_data, 'r', label="Signal", linewidth=3)
plt.plot(d_data, 'k.', label="Data") plt.plot(d_data, 'k.', label="Data")
plt.plot(m_data, 'k', label="Reconstruction",linewidth=3) plt.plot(m_data, 'k', label="Reconstruction",linewidth=3)
plt.title("Reconstruction") plt.title("Reconstruction")
plt.legend() plt.legend()
plt.show() plt.show()
``` ```
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
plt.figure(figsize=(15,10)) plt.figure(figsize=(15,10))
plt.plot(s_data - s_data, 'r', label="Signal", linewidth=3) plt.plot(s_data - s_data, 'r', label="Signal", linewidth=3)
plt.plot(d_data - s_data, 'k.', label="Data") plt.plot(d_data - s_data, 'k.', label="Data")
plt.plot(m_data - s_data, 'k', label="Reconstruction",linewidth=3) plt.plot(m_data - s_data, 'k', label="Reconstruction",linewidth=3)
plt.axhspan(-noise_amplitude,noise_amplitude, facecolor='0.9', alpha=.5) plt.axhspan(-noise_amplitude,noise_amplitude, facecolor='0.9', alpha=.5)
plt.title("Residuals") plt.title("Residuals")
plt.legend() plt.legend()
plt.show() plt.show()
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Power Spectrum ### Power Spectrum
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
s_power_data = ift.power_analyze(sh).to_global_data() s_power_data = ift.power_analyze(sh).to_global_data()
m_power_data = ift.power_analyze(m).to_global_data() m_power_data = ift.power_analyze(m).to_global_data()
plt.figure(figsize=(15,10)) plt.figure(figsize=(15,10))
plt.loglog() plt.loglog()
plt.xlim(1, int(N_pixels/2)) plt.xlim(1, int(N_pixels/2))
ymin = min(m_power_data) ymin = min(m_power_data)
plt.ylim(ymin, 1) plt.ylim(ymin, 1)
xs = np.arange(1,int(N_pixels/2),.1) xs = np.arange(1,int(N_pixels/2),.1)
plt.plot(xs, pow_spec(xs), label="True Power Spectrum", color='k',alpha=0.5) plt.plot(xs, pow_spec(xs), label="True Power Spectrum", color='k',alpha=0.5)
plt.plot(s_power_data, 'r', label="Signal") plt.plot(s_power_data, 'r', label="Signal")
plt.plot(m_power_data, 'k', label="Reconstruction") plt.plot(m_power_data, 'k', label="Reconstruction")
plt.axhline(noise_amplitude**2 / N_pixels, color="k", linestyle='--', label="Noise level", alpha=.5) plt.axhline(noise_amplitude**2 / N_pixels, color="k", linestyle='--', label="Noise level", alpha=.5)
plt.axhspan(noise_amplitude**2 / N_pixels, ymin, facecolor='0.9', alpha=.5) plt.axhspan(noise_amplitude**2 / N_pixels, ymin, facecolor='0.9', alpha=.5)
plt.title("Power Spectrum") plt.title("Power Spectrum")
plt.legend() plt.legend()
plt.show() plt.show()
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
## Wiener Filter on Incomplete Data ## Wiener Filter on Incomplete Data
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
# Operators # Operators
Sh = ift.create_power_operator(h_space, power_spectrum=pow_spec) Sh = ift.create_power_operator(h_space, power_spectrum=pow_spec)
N = ift.ScalingOperator(noise_amplitude**2,s_space) N = ift.ScalingOperator(noise_amplitude**2,s_space)
# R is defined below # R is defined below
# Fields # Fields
sh = Sh.draw_sample() sh = Sh.draw_sample()
s = HT(sh) s = HT(sh)
n = ift.Field.from_random(domain=s_space, random_type='normal', n = ift.Field.from_random(domain=s_space, random_type='normal',
std=noise_amplitude, mean=0) std=noise_amplitude, mean=0)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Partially Lose Data ### Partially Lose Data
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
l = int(N_pixels * 0.2) l = int(N_pixels * 0.2)
h = int(N_pixels * 0.2 * 2) h = int(N_pixels * 0.2 * 2)
mask = np.full(s_space.shape, 1.) mask = np.full(s_space.shape, 1.)
mask[l:h] = 0 mask[l:h] = 0
mask = ift.Field.from_global_data(s_space, mask) mask = ift.Field.from_global_data(s_space, mask)
R = ift.DiagonalOperator(mask)(HT) R = ift.DiagonalOperator(mask)(HT)
n = n.to_global_data_rw() n = n.to_global_data_rw()
n[l:h] = 0 n[l:h] = 0
n = ift.Field.from_global_data(s_space, n) n = ift.Field.from_global_data(s_space, n)
d = R(sh) + n d = R(sh) + n
``` ```
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
curv = Curvature(R=R, N=N, Sh=Sh) curv = Curvature(R=R, N=N, Sh=Sh)
D = curv.inverse D = curv.inverse
j = R.adjoint_times(N.inverse_times(d)) j = R.adjoint_times(N.inverse_times(d))
m = D(j) m = D(j)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Compute Uncertainty ### Compute Uncertainty
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
m_mean, m_var = ift.probe_with_posterior_samples(curv, HT, 200) m_mean, m_var = ift.probe_with_posterior_samples(curv, HT, 200)
``` ```
%% Cell type:markdown id: tags: %% Cell type:markdown id: tags:
### Get data ### Get data
%% Cell type:code id: tags: %% Cell type:code id: tags:
``` python ``` python
# Get signal data and reconstruction data # Get signal data and reconstruction data
s_data = s.to_global_data() s_data = s.to_global_data()
m_data = HT(m).to_global_data() m_data = HT(m).to_global_data()
m_var_data = m_var.to_global_data() m_var_data = m_var.to_global_data()