Regular update (#1047)

* update notebooks

* update notebooks

* update config

* update exp_record

* update MultiDetectEncoder

* update MultiDetectEncoder

* update notebooks; update SourceTypeAccuracy to report w.r.t magnitudes

* update notebooks

* add ellipticity estimation

* fix tests

* fix the warning in sampler

* add a new notebook to show the mse of ellipticity estimation

* change MSE to residual for ellipticity measure

* modify units in catalog; remove background from dc2

* replace on_fluxes, magnitudes, magnitudes_njy with on_nmgy, on_mag and on_njy

* fix notebook after merge

* change the gating logic of VariationalFactor

* delete old notebooks

* address pr comments

---------

Co-authored-by: Yicun Duan <[email protected]>

Author: YicunDuanUMich

bliss/cached_dataset.py

@@ -90,7 +90,7 @@ def __call__(self, datum_in):
 
 class ChunkingSampler(Sampler):
     def __init__(self, dataset: Dataset) -> None:
-        super().__init__(dataset)
+        super().__init__()
         assert isinstance(dataset, ChunkingDataset), "dataset should be ChunkingDataset"
         self.dataset = dataset
 

bliss/catalog.py

@@ -19,7 +19,22 @@ def convert_nmgy_to_mag(nmgy):
 
 
 def convert_nmgy_to_njymag(nmgy):
-    """Convert from flux (nano-maggie) to mag (nano-jansky), which is the format used by DC2."""
+    """Convert from flux (nano-maggie) to mag (nano-jansky), which is the format used by DC2.
+
+    For the difference between mag (Pogson magnitude) and njymag (AB magnitude), please view
+    the "Flux units: maggies and nanomaggies" part of
+    https://www.sdss3.org/dr8/algorithms/magnitudes.php#nmgy
+    When we change the standard source to AB sources, we need to do the conversion
+    described in "2.10 AB magnitudes" at
+    https://pstn-001.lsst.io/fluxunits.pdf
+
+    Args:
+        nmgy: the fluxes in nanomaggies
+
+    Returns:
+        Tensor indicating fluxes in AB magnitude
+    """
+
     return 22.5 - 2.5 * torch.log10(nmgy / 3631)
 
 
@@ -148,13 +163,19 @@ def galaxy_bools(self) -> Tensor:
         is_galaxy = self["source_type"] == SourceType.GALAXY
         return is_galaxy * self.is_on_mask.unsqueeze(-1)
 
-    @property
-    def on_fluxes(self):
-        """Gets fluxes of "on" sources based on whether the source is a star or galaxy.
+    def on_fluxes(self, unit: str):
+        match unit:
+            case "nmgy":
+                return self.on_nmgy
+            case "mag":
+                return self.on_mag
+            case "njymag":
+                return self.on_njymag
+            case _:
+                raise NotImplementedError()
 
-        Returns:
-            Tensor: a tensor of fluxes of size (b x nth x ntw x max_sources x 1)
-        """
+    @property
+    def on_nmgy(self):
         # TODO: a tile catalog should store fluxes rather than star_fluxes and galaxy_fluxes
         # because that's all that's needed to render the source
         if "galaxy_fluxes" not in self:
@@ -164,13 +185,12 @@ def on_fluxes(self):
         return torch.where(self.is_on_mask[..., None], fluxes, torch.zeros_like(fluxes))
 
     @property
-    def magnitudes(self):
-        # TODO: we shouldn't assume fluxes are stored in nanomaggies because they aren't for DC2
-        return convert_nmgy_to_mag(self.on_fluxes)
+    def on_mag(self) -> Tensor:
+        return convert_nmgy_to_mag(self.on_nmgy)
 
     @property
-    def magnitudes_njy(self):
-        return convert_nmgy_to_njymag(self.on_fluxes)
+    def on_njymag(self) -> Tensor:
+        return convert_nmgy_to_njymag(self.on_nmgy)
 
     def to_full_catalog(self, tile_slen):
         """Converts image parameters in tiles to parameters of full image.
@@ -266,8 +286,8 @@ def get_indices_of_on_sources(self) -> Tuple[Tensor, Tensor]:
 
     def _sort_sources_by_flux(self, band=2):
         # sort by fluxes of "on" sources to get brightest source per tile
-        on_fluxes = self.on_fluxes[..., band]  # shape n x nth x ntw x d
-        top_indexes = on_fluxes.argsort(dim=3, descending=True)
+        on_nmgy = self.on_nmgy[..., band]  # shape n x nth x ntw x d
+        top_indexes = on_nmgy.argsort(dim=3, descending=True)
 
         d = {"n_sources": self["n_sources"]}
         for key, val in self.items():
@@ -323,8 +343,8 @@ def filter_by_flux(self, min_flux=0, max_flux=torch.inf, band=2):
         sorted_self = self._sort_sources_by_flux(band=band)
 
         # get fluxes of "on" sources to mask by
-        on_fluxes = sorted_self.on_fluxes[..., band]
-        flux_mask = (on_fluxes > min_flux) & (on_fluxes < max_flux)
+        on_nmgy = sorted_self.on_nmgy[..., band]
+        flux_mask = (on_nmgy > min_flux) & (on_nmgy < max_flux)
 
         d = {}
         for key, val in sorted_self.items():
@@ -458,13 +478,19 @@ def galaxy_bools(self) -> Tensor:
         assert is_galaxy.size(2) == 1
         return is_galaxy * self.is_on_mask.unsqueeze(2)
 
-    @property
-    def on_fluxes(self) -> Tensor:
-        """Gets fluxes of "on" sources based on whether the source is a star or galaxy.
+    def on_fluxes(self, unit: str):
+        match unit:
+            case "nmgy":
+                return self.on_nmgy
+            case "mag":
+                return self.on_mag
+            case "njymag":
+                return self.on_njymag
+            case _:
+                raise NotImplementedError()
 
-        Returns:
-            Tensor: a tensor of fluxes of size (b x nth x ntw x max_sources x 1)
-        """
+    @property
+    def on_nmgy(self) -> Tensor:
         # ideally we'd always store fluxes rather than star_fluxes and galaxy_fluxes
         if "galaxy_fluxes" not in self:
             fluxes = self["star_fluxes"]
@@ -473,8 +499,12 @@ def on_fluxes(self) -> Tensor:
         return torch.where(self.is_on_mask[..., None], fluxes, torch.zeros_like(fluxes))
 
     @property
-    def magnitudes(self):
-        return convert_nmgy_to_mag(self.on_fluxes)
+    def on_mag(self) -> Tensor:
+        return convert_nmgy_to_mag(self.on_nmgy)
+
+    @property
+    def on_njymag(self) -> Tensor:
+        return convert_nmgy_to_njymag(self.on_nmgy)
 
     def one_source(self, b: int, s: int):
         """Return a dict containing all parameter for one specified light source."""

bliss/conf/base_config.yaml

@@ -80,64 +80,75 @@ variational_factors:
       name: locs
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 d -> b ht wt d"
-      nll_gating: n_sources
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.SourcesGating
     - _target_: bliss.encoder.variational_dist.BernoulliFactor
       name: source_type
       sample_rearrange: "b ht wt -> b ht wt 1 1"
       nll_rearrange: "b ht wt 1 1 -> b ht wt"
-      nll_gating: n_sources
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.SourcesGating
     - _target_: bliss.encoder.variational_dist.LogNormalFactor
       name: star_fluxes
       dim: 5
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 d -> b ht wt d"
-      nll_gating: is_star
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.StarGating
     - _target_: bliss.encoder.variational_dist.LogNormalFactor
       name: galaxy_fluxes
       dim: 5
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 d -> b ht wt d"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
     - _target_: bliss.encoder.variational_dist.LogitNormalFactor
       name: galaxy_disk_frac
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 1 -> b ht wt 1"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
     - _target_: bliss.encoder.variational_dist.LogitNormalFactor
       name: galaxy_beta_radians
       high: 3.1415926
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 1 -> b ht wt 1"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
     - _target_: bliss.encoder.variational_dist.LogitNormalFactor
       name: galaxy_disk_q
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 1 -> b ht wt 1"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
     - _target_: bliss.encoder.variational_dist.LogNormalFactor
       name: galaxy_a_d
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 1 -> b ht wt 1"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
     - _target_: bliss.encoder.variational_dist.LogitNormalFactor
       name: galaxy_bulge_q
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 1 -> b ht wt 1"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
     - _target_: bliss.encoder.variational_dist.LogNormalFactor
       name: galaxy_a_b
       sample_rearrange: "b ht wt d -> b ht wt 1 d"
       nll_rearrange: "b ht wt 1 1 -> b ht wt 1"
-      nll_gating: is_galaxy
+      nll_gating:
+        _target_: bliss.encoder.variational_dist.GalaxyGating
 
 metrics:
     detection_performance:
         _target_: bliss.encoder.metrics.DetectionPerformance
-        mag_bin_cutoffs: [19, 19.4, 19.8, 20.2, 20.6, 21, 21.4, 21.8]
-        mag_band: 2
+        bin_cutoffs: [23.9, 24.1, 24.5, 24.9, 25.6]
+        bin_type: "njymag"
     source_type_accuracy:
         _target_: bliss.encoder.metrics.SourceTypeAccuracy
-        flux_bin_cutoffs: [200, 400, 600, 800, 1000]
+        bin_cutoffs: [23.9, 24.1, 24.5, 24.9, 25.6]
+        bin_type: "njymag"
     flux_error:
         _target_: bliss.encoder.metrics.FluxError
         survey_bands: ${encoder.survey_bands}
@@ -230,7 +241,7 @@ surveys:
         n_image_split: 50
         tile_slen: 4
         max_sources_per_tile: 5
-        min_flux: 0.0
+        catalog_min_r_flux: 50
         prepare_data_processes_num: 4
         data_in_one_cached_file: 1250
         splits: 0:80/80:90/90:100
@@ -242,7 +253,13 @@ surveys:
             - _target_: bliss.data_augmentation.RandomShiftTransform
               tile_slen: ${surveys.dc2.tile_slen}
               max_sources_per_tile: ${surveys.dc2.tile_slen}
-        nontrain_transforms: []
+            - _target_: bliss.cached_dataset.FluxFilterTransform
+              reference_band: 2  # r-band
+              min_flux: 100
+        nontrain_transforms:
+            - _target_: bliss.cached_dataset.FluxFilterTransform
+              reference_band: 2  # r-band
+              min_flux: 100
 
 
 #######################################################################

bliss/encoder/encoder.py

@@ -26,7 +26,7 @@ def __init__(
         self,
         survey_bands: list,
         tile_slen: int,
-        image_normalizers: list,
+        image_normalizers: dict,
         var_dist: VariationalDist,
         matcher: CatalogMatcher,
         sample_image_renders: MetricCollection,
@@ -105,7 +105,7 @@ def make_context(self, history_cat, history_mask, detection2=False):
             )
         else:
             centered_locs = history_cat["locs"][..., 0, :] - 0.5
-            log_fluxes = (history_cat.on_fluxes.squeeze(3).sum(-1) + 1).log()
+            log_fluxes = (history_cat.on_nmgy.squeeze(3).sum(-1) + 1).log()
             history_encoding_lst = [
                 history_cat["n_sources"].float(),  # detection history
                 log_fluxes * history_cat["n_sources"],  # flux history