Revised CalCat API

This is based on our discussion last week, but it's still a WIP. In particular, I haven't implemented actually loading the constants yet.

I've tweaked the design slightly: rather than having CalibrationData subclasses for each detector type, you use CalibrationData.from_conditions() with a detector specific object like AGIPDConditions. Jump to the tests to see what this looks like.

Questions:

• How to deal with missing modules?
  • A module may be missing for one type of constant but not for another
  • How do we even tell if modules are missing - should we query the PDUs so we know how many modules to expect?
• How to deal with missing calibration types (e.g. SlopesPC)
  • Especially if one group (e.g. illuminated) not found - this currently raises an error, even if we have other constants we could return
• Naming:
  • 'Calibrations' vs 'calibration' types vs 'constant types' (e.g. Offset, SlopesPC
  • 'Modules' vs 'aggregators'
• How many of SingleConstantVersion fields do we actually want?
• Should .select_modules() and .select_calibrations() take *args or a list?

cc @schmidtp @ahmedk @hammerd

src/cal_tools/calcat_interface2.py

@@ -218,6 +218,31 @@ class SingleConstantVersion:
         return self.dataset_obj(caldb_root)[:]
 
 
+def prepare_selection(
+    module_details, module_nums=None, aggregator_names=None, qm_names=None
+):
+    aggs = aggregator_names  # Shorter name -> fewer multi-line statements
+    n_specified = sum([module_nums is not None, aggs is not None, qm_names is not None])
+    if n_specified > 1:
+        raise TypeError(
+            "select_modules() accepts only one of module_nums, aggregator_names & qm_names"
+        )
+
+    if module_nums is not None:
+        by_mod_no = {m["module_number"]: m for m in module_details}
+        return [by_mod_no[n]["karabo_da"] for n in module_nums]
+    elif qm_names is not None:
+        by_qm = {m["virtual_device_name"]: m for m in module_details}
+        return [by_qm[s]["karabo_da"] for s in qm_names]
+    elif aggs is not None:
+        miss = set(aggs) - {m["karabo_da"] for m in module_details}
+        if miss:
+            raise KeyError("Aggregators not found: " + ", ".join(sorted(miss)))
+        return aggs
+    else:
+        raise TypeError("select_modules() requires an argument")
+
+
 @dataclass
 class ModulesConstantVersions:
     """A group of similar CCVs for several modules of one detector"""
@@ -226,36 +251,19 @@ class ModulesConstantVersions:
     module_details: List[Dict]
 
     def select_modules(
-        self, module_nums=None, *, aggregators=None, qm_names=None
+        self, module_nums=None, *, aggregator_names=None, qm_names=None
     ) -> "ModulesConstantVersions":
-        n_specified = sum(
-            [module_nums is not None, aggregators is not None, qm_names is not None]
+        aggs = prepare_selection(
+            self.module_details, module_nums, aggregator_names, qm_names
         )
-        if n_specified < 1:
-            raise TypeError("select_modules() requires an argument")
-        elif n_specified > 1:
-            raise TypeError(
-                "select_modules() accepts only one of module_nums, aggregators & qm_names"
-            )
-
-        if module_nums is not None:
-            by_mod_no = {m["module_number"]: m for m in self.module_details}
-            aggregators = [by_mod_no[n]["karabo_da"] for n in module_nums]
-        elif qm_names is not None:
-            by_qm = {m["virtual_device_name"]: m for m in self.module_details}
-            aggregators = [by_qm[s]["karabo_da"] for s in qm_names]
-        elif aggregators is not None:
-            miss = set(aggregators) - {m["karabo_da"] for m in self.module_details}
-            if miss:
-                raise KeyError("Aggregators not found: " + ", ".join(sorted(miss)))
-
-        d = {aggr: scv for (aggr, scv) in self.constants.items() if aggr in aggregators}
-        return ModulesConstantVersions(d, self.module_details)
+        d = {aggr: scv for (aggr, scv) in self.constants.items() if aggr in aggs}
+        mods = [m for m in self.module_details if m["karabo_da"] in d]
+        return ModulesConstantVersions(d, mods)
 
     # These properties label only the modules we have constants for, which may
     # be a subset of what's in module_details
     @property
-    def aggregators(self):
+    def aggregator_names(self):
         return sorted(self.constants)
 
     @property
@@ -274,33 +282,44 @@ class ModulesConstantVersions:
             if m["karabo_da"] in self.constants
         ]
 
-    def ndarray(self, caldb_root=None):
-        eg_dset = self.constants[self.aggregators[0]].dataset_obj(caldb_root)
+    def ndarray(self, caldb_root=None, *, parallel=0):
+        eg_dset = self.constants[self.aggregator_names[0]].dataset_obj(caldb_root)
         shape = (len(self.constants),) + eg_dset.shape
-        arr = np.zeros(shape, eg_dset.dtype)
-        for i, agg in enumerate(self.aggregators):
-            dset = self.constants[agg].dataset_obj(caldb_root)
-            dset.read_direct(arr[i])
+
+        if parallel > 0:
+            load_ctx = psh.ProcessContext(num_workers=parallel)
+        else:
+            load_ctx = psh.SerialContext()
+
+        arr = psh.alloc(shape, eg_dset.dtype, fill=0)
+
+        def _load_constant_dataset(wid, index, mod):
+            dset = self.constants[mod].dataset_obj(caldb_root)
+            dset.read_direct(arr[index])
+
+        load_ctx.map(_load_constant_dataset, self.aggregator_names)
         return arr
 
-    def xarray(self, module_naming="da", caldb_root=None):
+    def xarray(self, module_naming="modnum", caldb_root=None, *, parallel=0):
         import xarray
 
-        if module_naming == "da":
-            modules = self.aggregators
-        elif module_naming == "modno":
+        if module_naming == "aggregator":
+            modules = self.aggregator_names
+        elif module_naming == "modnum":
             modules = self.module_nums
         elif module_naming == "qm":
             modules = self.qm_names
         else:
-            raise ValueError(f"{module_naming=} (must be 'da', 'modno' or 'qm'")
+            raise ValueError(
+                f"{module_naming=} (must be 'aggregator', 'modnum' or 'qm'"
+            )
 
-        ndarr = self.ndarray(caldb_root)
+        ndarr = self.ndarray(caldb_root, parallel=parallel)
 
         # Dimension labels
         dims = ["module"] + ["dim_%d" % i for i in range(ndarr.ndim - 1)]
         coords = {"module": modules}
-        name = self.constants[self.aggregators[0]].constant_name
+        name = self.constants[self.aggregator_names[0]].constant_name
 
         return xarray.DataArray(ndarr, dims=dims, coords=coords, name=name)
 
@@ -340,10 +359,7 @@ class CalibrationData(Mapping):
     """Collected constants for a given detector"""
 
     def __init__(self, constant_groups, module_details):
-        self.constant_groups = {
-            const_type: ModulesConstantVersions(d, module_details)
-            for const_type, d in constant_groups.items()
-        }
+        self.constant_groups = constant_groups
         self.module_details = module_details
 
     @staticmethod
@@ -398,7 +414,7 @@ class CalibrationData(Mapping):
             if mod.get("module_number", -1) < 0:
                 mod["module_number"] = int(re.findall(r"\d+", mod["karabo_da"])[-1])
 
-        d = {}
+        constant_groups = {}
 
         for params, cal_types in cal_types_by_params_used.items():
             condition_dict = condition.make_dict(params)
@@ -424,11 +440,14 @@ class CalibrationData(Mapping):
                 aggr = ccv["physical_detector_unit"]["karabo_da"]
                 cal_type = cal_id_map[ccv["calibration_constant"]["calibration_id"]]
 
-                d.setdefault(cal_type, {})[aggr] = SingleConstantVersion.from_response(
-                    ccv
-                )
+                const_group = constant_groups.setdefault(cal_type, {})
+                const_group[aggr] = SingleConstantVersion.from_response(ccv)
 
-        return cls(d, module_details)
+        mcvs = {
+            const_type: ModulesConstantVersions(d, module_details)
+            for const_type, d in constant_groups.items()
+        }
+        return cls(mcvs, module_details)
 
     @classmethod
     def from_report(
@@ -447,16 +466,22 @@ class CalibrationData(Mapping):
 
         res = client.get("calibration_constant_versions", params)
 
-        d = {}
-        pdus = set()
+        constant_groups = {}
+        pdus = []
 
         for ccv in res:
-            pdus.add(ccv["physical_detector_unit"])
+            pdus.append(ccv["physical_detector_unit"])
             cal_type = calibration_name(ccv["calibration_constant"]["calibration_id"])
             aggr = ccv["physical_detector_unit"]["karabo_da"]
-            d.setdefault(cal_type, {})[aggr] = SingleConstantVersion.from_response(ccv)
+            const_group = constant_groups.setdefault(cal_type, {})
+            const_group[aggr] = SingleConstantVersion.from_response(ccv)
 
-        return cls(d, sorted(pdus, key=lambda d: d["karabo_da"]))
+        module_details = sorted(pdus, key=lambda d: d["karabo_da"])
+        mcvs = {
+            const_type: ModulesConstantVersions(d, module_details)
+            for const_type, d in constant_groups.items()
+        }
+        return cls(mcvs, module_details)
 
     def __getitem__(self, key) -> ModulesConstantVersions:
         return self.constant_groups[key]
@@ -494,39 +519,28 @@ class CalibrationData(Mapping):
 
     def require_calibrations(self, calibrations):
         """Drop any modules missing the specified constant types"""
-        mods = set(self.aggregators)
+        mods = set(self.aggregator_names)
         for cal_type in calibrations:
             mods.intersection_update(self[cal_type].constants)
-        return self.select_modules(mods)
-
-    def select_calibrations(self, calibrations, require_all=True):
-        if require_all:
-            missing = set(calibrations) - set(self.constant_groups)
-            if missing:
-                raise KeyError(f"Missing calibrations: {', '.join(sorted(missing))}")
-
-        d = {
-            cal_type: mcv.constants
-            for (cal_type, mcv) in self.constant_groups.items()
-            if cal_type in calibrations
-        }
-        # TODO: missing for some modules?
-        return type(self)(d, self.aggregators)
+        return self.select_modules(aggregator_names=mods)
 
     def select_modules(
-        self, module_nums=None, *, aggregators=None, qm_names=None
+        self, module_nums=None, *, aggregator_names=None, qm_names=None
     ) -> "CalibrationData":
-        return type(self)(
-            {
-                cal_type: mcv.select_modules(
-                    module_nums=module_nums,
-                    aggregators=aggregators,
-                    qm_names=qm_names,
-                ).constants
-                for (cal_type, mcv) in self.constant_groups.items()
-            },
-            sorted(aggregators),
+        # Validate the specified modules against those we know about.
+        # Each specific constant type may have only a subset of these modules.
+        aggs = prepare_selection(
+            self.module_details, module_nums, aggregator_names, qm_names
         )
+        mcvs = {
+            cal_type: mcv.select_modules(
+                aggregator_names=set(aggs).intersection(mcv.aggregator_names)
+            )
+            for (cal_type, mcv) in self.constant_groups.items()
+        }
+        aggs = set().union(*[c.aggregator_names for c in mcvs.values()])
+        module_details = [m for m in self.module_details if m["karabo_da"] in aggs]
+        return type(self)(mcvs, module_details)
 
     def merge(self, *others: "CalibrationData") -> "CalibrationData":
         d = {}
@@ -536,34 +550,12 @@ class CalibrationData(Mapping):
                 if cal_type in other:
                     d[cal_type].update(other[cal_type].constants)
 
-        aggregators = set(self.aggregators)
+        aggregators = set(self.aggregator_names)
         for other in others:
-            aggregators.update(other.aggregators)
+            aggregators.update(other.aggregator_names)
 
         return type(self)(d, sorted(aggregators))
 
-    def load_all(self, caldb_root=None):
-        res = {}
-
-        const_load_mp = psh.ProcessContext(num_workers=24)
-        keys = []
-        for cal_type, mcv in self.constant_groups.items():
-            res[cal_type] = {}
-            for module in mcv.aggregators:
-                dset = mcv.constants[module].dataset_obj(caldb_root)
-                res[cal_type][module] = const_load_mp.alloc(
-                    shape=dset.shape, dtype=dset.dtype
-                )
-                keys.append((cal_type, module))
-
-        def _load_constant_dataset(wid, index, key):
-            cal_type, mod = key
-            dset = self[cal_type].constants[mod].dataset_obj(caldb_root)
-            dset.read_direct(res[cal_type][mod])
-
-        const_load_mp.map(_load_constant_dataset, keys)
-        return res
-
 
 class ConditionsBase:
     calibration_types = {}  # For subclasses: {calibration: [parameter names]}