Changes made to make mogi and cdm run properly by gracebato · Pull Request #11 · AlTarFramework/altar

gracebato · 2019-03-24T06:13:14Z

proper calculation of LLK (e.g. LLK = normalization - normeval**2.0 / 2.0 )
avoiding 0 values in the random generator within Metropolis.py
dV is now treated as a non-Jeffrey parameter to allow negative source (deflation).
LOS matrix for the EW component should have a negative sign.

… 2) avoiding 0 values in the random generator within Metropolis.py, 3) dV is treated as a non-Jeffrey parameter.

lijun99 · 2019-03-24T18:19:44Z

The framework changes were also included in my previous pull request.

proper calculation of LLK (e.g. LLK = normalization - normeval**2.0 / 2.0 )
I have proposed to use ddot (v*v) instead of dnrm2 (|v|) in L2.py:eval. Maybe a compromise is that we define a new function evalsquare for (v*v) so that users can pick either eval (LLK = -0.5*norm*norm) or evalsquare (LLK = -0.5*normsq) in their models. I will test which one is more efficient, but judging from openblas x86_64 kernels, ddot is better optimized with streaming SIMD extensions.
avoiding 0 values in the random generator within Metropolis.py
gsl offers a positive uniform distribution implementation, which is now included in pyre. Use
self.uniform = altar.pdf.uniform_pos(rng=rng) instead. Adding a range requires rescaling for each generated random number, which might not be efficient.

gracebato · 2019-04-25T22:23:45Z

models/cdm/cdm/Fast.py

        libcdm.layout(source,
-                      model.xIdx, model.dIdx,
-                      model.openingIdx, model.aXIdx, model.omegaXIdx,
+                      model.xIdx, model.dIdx, model.openingIdx, model.aXIdx, model.omegaXIdx,


Modified function variable referencing/order for consistency.

this looks like merging two lines into one; am i missing something

gracebato · 2019-04-25T22:24:30Z

models/cdm/cdm/Fast.py

            # compute the norm, and normalize it
-            llk = normalization - norm.eval(v=residuals, sigma_inv=cd_inv) / 2
+            normeval = norm.eval(v=residuals, sigma_inv=cd_inv)
+            llk = normalization - normeval**2.0 / 2.0


Corrected the calculation of LLK.

good catch!

gracebato · 2019-04-25T22:25:58Z

models/cdm/cdm/Native.py

        xIdx = model.xIdx
        yIdx = model.yIdx
        dIdx = model.dIdx
+        openingIdx = model.openingIdx


Nothing significant. Just reordered the variable for consistency.

gracebato · 2019-04-25T22:26:19Z

models/cdm/cdm/Native.py


            # make a source using the sample parameters
-            cdm = source(x=x, y=y, d=d,
+            cdm = source(x=x, y=y, d=d, opening=opening,


Modified function variable referencing/order for consistency.

with python pass-by-name, the order of arguments is not important

gracebato · 2019-04-25T22:27:16Z

models/cdm/cdm/Native.py

+            normeval = norm.eval(v=u, sigma_inv=cd_inv)
            # normalize and store it as the data log likelihood
-            dataLLK[sample] = normalization - residual/2
+            dataLLK[sample] = normalization - normeval**2.0 /2


Corrected the calculation of the LLK.

gracebato · 2019-04-25T22:29:16Z

models/cdm/cdm/Source.py

-        ue, un, uv =  CDM(Xf, Yf, x_src, y_src, d_src,
-                          omegaX_src, omegaY_src, omegaZ_src, ax_src, ay_src, az_src,
-                          opening, v)
+        ue, un, uv =  CDM(Xf, Yf, x_src, y_src, d_src, opening,


Lines 81-83: Modified function variable referencing/order for consistency.

gracebato · 2019-04-25T22:29:41Z

models/cdm/cdm/Source.py

    # meta-methods
-    def __init__(self, x=x, y=y, d=d, omegaX=omegaX, omegaY=omegaY, omegaZ=omegaZ,
-                 ax=ax, ay=ay, az=az, opening=opening, v=v, **kwds):
+    def __init__(self, x=x, y=y, d=d, opening=opening, ax=ax, ay=ay, az=az,


Lines 95-96: Same here--modified function variable referencing/order for consistency.

gracebato · 2019-04-25T22:29:56Z

models/cdm/cdm/libcdm.py



-def CDM(X, Y, X0, Y0, depth, omegaX, omegaY, omegaZ, ax, ay, az, opening, nu):
+def CDM(X, Y, X0, Y0, depth, opening, ax, ay, az, omegaX, omegaY, omegaZ, nu):


Also here--modified function variable referencing/order for consistency.

gracebato · 2019-04-25T22:30:58Z

models/cdm/examples/synthetic/cdm.py

-        import itertools
-        # build a set of points on a grid
-        stations = itertools.product(range(-5,6), range(-5,6))
+        import numpy as np


Line 171-177: I think for the synthetic case, this is better to constrain the number of points.

gracebato · 2019-04-25T22:41:21Z

models/cdm/examples/synthetic/cdm.py

        # make a source
        source = altar.models.cdm.source(
-            x=self.x, y=self.y, d=self.d,
+            x=self.x, y=self.y, d=self.d, opening=self.opening,


Lines 99-102: Modified function variable referencing/order for consistency.

…aset.

gracebato · 2019-04-25T22:45:33Z

models/cdm/examples/synthetic/cdm.py

        for idx, observation in enumerate(data):
            # set the covariance to a fraction of the "observed" displacement
-            correlation[idx,idx] = 1.0 #.01 * observation.d
+            correlation[idx,idx] = 0.0001 #1.0 #.01 * observation.d


1.0 m is quite large for an observation error (variance).

gracebato · 2019-04-25T22:46:28Z

models/cdm/lib/libcdm/Source.cc

-        // compute the displacements
        cdm(_locations,
            xSrc, ySrc, dSrc,
+            openingSrc,


Modified function variable referencing/order for consistency.

gracebato · 2019-04-25T22:47:23Z

models/cdm/lib/libcdm/Source.cc

+            auto uy = gsl_matrix_get(disp, loc, 1);
+            auto ud = gsl_matrix_get(disp, loc, 2);

-            // project; don't forget {ud} is positive into the ground


This is not a depth value. It is the vertical component.

gracebato · 2019-04-25T22:48:16Z

models/cdm/lib/libcdm/Source.cc

-            // project; don't forget {ud} is positive into the ground
-            auto u = ux*nx + uy*ny - ud*nz;
+            // project
+            auto u = ux*nx + uy*ny + ud*nz;


This is not a depth value and should be '+' sign.

gracebato · 2019-04-25T22:48:39Z

models/cdm/lib/libcdm/cdm.cc

+    double opening,
    double aX, double aY, double aZ,
    double omegaX, double omegaY, double omegaZ,
-    double opening,


Modified function variable referencing/order for consistency.

gracebato · 2019-04-25T22:52:39Z

models/cdm/lib/libcdm/cdm.h

        namespace cdm {
            void cdm(const gsl_matrix * locations,
                     double X0, double Y0, double depth,
+                     double opening,


Modified function variable referencing/order for consistency.

gracebato added 2 commits March 22, 2019 19:08

Changes made to make mogi run properly: 1) proper calculation of LLK,…

4a29dfc

… 2) avoiding 0 values in the random generator within Metropolis.py, 3) dV is treated as a non-Jeffrey parameter.

modified 1.0e6** to 1.e06*

cc66c00

aivazis self-assigned this Apr 11, 2019

Corrections to make CDM work properly.

7dc60a1

gracebato commented Apr 25, 2019

View reviewed changes

gracebato added 3 commits April 25, 2019 15:34

Removed print function that I used when validating the code.

c108ede

Removed print lines that I used during code validation.

a4bb6cc

Removed 'h' in Line 1. Typo error.

d86d3cb

gracebato commented Apr 25, 2019

View reviewed changes

Removed separation of E and W oids. You cannot do this within one dat…

d42abb0

…aset.

gracebato commented Apr 25, 2019

View reviewed changes

gracebato changed the title ~~Changes made to make mogi run properly~~ Changes made to make mogi and cdm run properly Apr 25, 2019

aivazis added a commit that referenced this pull request May 31, 2019

models/cdm: applied fixes from #11 submitted by @gracebato

ac67ad0

gracebato pushed a commit to gracebato/altar that referenced this pull request Jul 15, 2019

models/cdm: applied fixes from AlTarFramework#11 submitted by @gracebato

672984d



		def CDM(X, Y, X0, Y0, depth, omegaX, omegaY, omegaZ, ax, ay, az, opening, nu):
		def CDM(X, Y, X0, Y0, depth, opening, ax, ay, az, omegaX, omegaY, omegaZ, nu):

Conversation

gracebato commented Mar 24, 2019

Uh oh!

lijun99 commented Mar 24, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gracebato Apr 25, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gracebato Apr 25, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gracebato Apr 25, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

lijun99 commented Mar 24, 2019 •

edited

Loading

gracebato Apr 25, 2019 •

edited

Loading

gracebato Apr 25, 2019 •

edited

Loading

gracebato Apr 25, 2019 •

edited

Loading