Correlated observation (#239)

* added mask to CorrelatedObservation; works now for multi-band data

* much faster correlation function

* first stab at resampling observations

* AsinhAutomaticNorm works with CorrelatedObservation

* removed automatic obs.match(model_frame) in Frame.from_observations; refactored renderer

* set weight according to power spectrum; fix padding

* renderer set on demand

* simplified test

* docs update: resampling for multi-res fits

* getting rid of ruff complaints (hopefully)

* corrected model colors in plot.sources

* added Observation.name as consistent plot label

Author: pmelchior

docs/0-quickstart.ipynb

@@ -84,12 +84,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "obs = sc2.Observation(\n",
-    "    data,\n",
-    "    weights,\n",
-    "    psf=sc2.ArrayPSF(psf),\n",
-    "    channels=channels,\n",
-    ")"
+    "obs = sc2.Observation(data, weights=weights, psf=psf, channels=channels)"
    ]
   },
   {
@@ -168,7 +163,7 @@
     "To fit multiple images with different PSF, different bands, different resolutions, etc..., we must define a model that is superior in all of these aspects, so that each observation can be computed from the model by a degradation operation, the so-called \"forward operator\". The point of joint modeling is to find one superior model of the sky that is consistent with all observations.\n",
     "```\n",
     "\n",
-    "In _scarlet2_, the forward operators are implemented as {py:class}`~scarlet2.renderer.Renderer`, which contains, e.g., PSF difference kernels and resampling transformations. It was created earlier by {py:func}`~scarlet2.Frame.from_observations`."
+    "In _scarlet2_, the forward operators are implemented as {py:class}`~scarlet2.renderer.Renderer`, which contains, e.g., PSF difference kernels and resampling transformations. It gets created when calling `~scarlet2.Observation.match` (which is normally done only when needed, but we can call it explicitly to see its effect):"
    ]
   },
   {
@@ -177,7 +172,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "obs.renderer"
+    "obs.match(model_frame)\n",
+    "print(obs.renderer)"
    ]
   },
   {

docs/howto/correlated.ipynb

@@ -97,9 +97,10 @@
    "source": [
     "# Scarlet Observations\n",
     "obs_hst = sc2.Observation(\n",
-    "    data_hst, wcs=wcs_hst, psf=sc2.ArrayPSF(psf_hst_data), channels=[\"F814W\"], weights=obs_hst_weights\n",
+    "    data_hst, wcs=wcs_hst, psf=psf_hst_data, channels=[\"F814W\"], weights=obs_hst_weights,\n",
     ")\n",
     "model_frame = sc2.Frame.from_observations(observations=obs_hst)\n",
+    "obs_hst.match(model_frame)\n",
     "norm_hst = sc2.plot.AsinhAutomaticNorm(obs_hst)\n",
     "sc2.plot.observation(obs_hst, sky_coords=ra_dec, label_kwargs={\"color\": \"red\"});"
    ]
@@ -144,7 +145,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "obs_corr = sc2.CorrelatedObservation.from_observation(obs_hst)"
+    "obs_corr = sc2.CorrelatedObservation.from_observation(obs_hst)\n",
+    "obs_corr.match(model_frame)"
    ]
   },
   {
@@ -269,7 +271,7 @@
    "id": "49d4d0ed0bc2467f",
    "metadata": {},
    "source": [
-    "We can see that the model picks up a lot of noise fluctuations. While one cannot rule out that some of them correspond to actual source emission, the residuals on the right show that the model is overfitting: the structure in the region of source 5 and 6 is much flatter (and importantly: shows smaller scale residuals) than in other regions.\n",
+    "A pretty decent fit, but there are some problems left. The model is missing flux around source 10 and the group of sources 0-4 in the left because the boxes are too narrow. We can also see that the model picks up a lot of noise fluctuations. While one cannot rule out that some of them correspond to actual source emission, the residuals on the right show that the model is overfitting: the structure in the region of source 5 and 6 is much flatter (and importantly: shows smaller scale residuals) than in other regions.\n",
     "\n",
     "Now we repeat the same process with the {py:class}`~scarlet2.CorrelatedObservation` we created earlier:"
    ]
@@ -308,7 +310,12 @@
   {
    "cell_type": "markdown",
    "id": "dc6f41d9437a271b",
-   "metadata": {},
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2025-12-02T18:59:38.199944Z",
+     "start_time": "2025-12-02T18:59:38.193610Z"
+    }
+   },
    "source": [
     "The residuals are more \"grainy\" now and look more consistent with the noise in other areas. In the region of sources 5 and 6, the model is still overfitting, but that's because we fit two sources to a single-band image. Such a fit is underconstrained and therefore prone to overfitting. Using additional observations in other filters can help (provided that sources 5 and 6 have different SEDs). See [the multi-observation tutorial](howto/multiresolution) for details."
    ]