Merge pull request #12 from KIPAC/dev

Dev
KIPAC · Aug 6, 2021 · b9b5811 · b9b5811
2 parents 112edfe + 4258edf
commit b9b5811
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 4 deletions.
diff --git a/docs/Week5_after.md b/docs/Week5_after.md
@@ -1,6 +1,6 @@
-# Follow up notes for week 4..
+# Follow up notes for week 5..
 
-### Model fitting and chi-squared.
+### Using code optimizers to do model fitting.
 
 Two weeks ago we learned about how to characterize the significance of data using
 a Gaussian distribution.

diff --git a/docs/Week6_after.md b/docs/Week6_after.md
@@ -4,7 +4,9 @@
 
 Two weeks ago, we learned how the `chi**2` could be used as a metric for the level of agreement between a model and data.
 
-Last week, we learned a few ways that using this `chi**2` may return less-than-ideal results. If parameters aren't chosen in a thoughtful way, then you may end up with correlated parameters, where after changing one, you must also change another to locally minimize the `chi**2`.
+Last week, we learned a few ways that using this `chi**2` may return less-than-ideal results. If parameters aren't chosen in a thoughtful way, then you may end up with correlated parameters, where after changing one, you must also change another to locally minimize the `chi**2`.  This makes it harder to describe and interpret the results because you can no longer consider the parameters individually.
+
+<img src=correlated_params.png alt="Correlated parameters" width="500"/>
 
 Moreover, we learned about some fitter idiosyncrasies. For example, if a reasonable number of parameters are chosen, and initial values are chosen close enough to the desired ones, then fitters can do a very nice job matching model to data:
 
@@ -20,4 +22,4 @@ The last idiosyncrasy we learned about was the fitter's ability to either not co
 
 <img src="NonConvergence.png" alt="Non-convergence of model to data" width="500"/>
 
-One good rule of thumb with fitting: always make sure to visually check how good your fit is - don't just rely on the magic of the minimizer and assume your fit will come out with sensible parameter values!
+One good rule of thumb with fitting: always make sure to visually check how good your fit is - don't just rely on the magic of the minimizer and assume your fit will come out with sensible parameter values!
diff --git a/docs/correlated_params.png b/docs/correlated_params.png