def run(self):
logger.info("TagsBuilder starting...")
self._build_indexes()
logger.info("Initializing list of all new task_ids to process ...")
previous_task_ids = []
for m in self._materials.find({"_tagsbuilder": {"$exists": True}},
{"_tagsbuilder.all_task_ids": 1}):
previous_task_ids.extend(m["_tagsbuilder"]["all_task_ids"])
previous_task_ids = [dbid_to_int(t) for t in previous_task_ids]
q = {"tags": {"$exists": True}, "task_id": {"$nin": previous_task_ids},
"state": "successful"}
tasks = [t for t in self._tasks.find(q, {"task_id": 1, "tags": 1})]
pbar = tqdm(tasks)
for t in pbar:
try:
pbar.set_description("Processing task_id: {}".format(t['task_id']))
# get the corresponding materials id
m = self._materials.find_one({"_tasksbuilder.all_task_ids":
dbid_to_str(self._tasks_prefix, t["task_id"])},
{"material_id": 1, "tags": 1,
"_tagsbuilder": 1})
if m:
all_tags = t["tags"]
if "tags" in m and m["tags"]:
all_tags.extend(m["tags"])
all_tasks = [dbid_to_str(self._tasks_prefix, t["task_id"])]
if "_tagsbuilder" in m:
all_tasks.extend(m["_tagsbuilder"]["all_task_ids"])
all_tags = list(set(all_tags)) # filter duplicates
self._materials.update_one({"material_id": m["material_id"]},
{"$set": {"tags": all_tags,
"_tagsbuilder.all_task_ids": all_tasks}})
except:
import traceback
logger.exception("<---")
logger.exception("There was an error processing task_id: {}".format(t["task_id"]))
logger.exception(traceback.format_exc())
logger.exception("--->")
logger.info("TagsBuilder finished processing.")
def run(self):
logger.info("MaterialsTaskBuilder starting...")
logger.info("Initializing list of all new task_ids to process ...")
previous_task_ids = []
for m in self._materials.find({}, {"_tasksbuilder.all_task_ids": 1}):
previous_task_ids.extend(m["_tasksbuilder"]["all_task_ids"])
q = {
"state": "successful",
"task_label": {
"$in": self.supported_task_labels
}
}
if self.query:
common_keys = [k for k in q.keys() if k in self.query.keys()]
if common_keys:
raise ValueError(
"User query parameter cannot contain key(s): {}".format(
common_keys))
q.update(self.query)
all_task_ids = [
dbid_to_str(self._t_prefix, t["task_id"])
for t in self._tasks.find(q, {"task_id": 1})
]
task_ids = [
t_id for t_id in all_task_ids if t_id not in previous_task_ids
]
logger.info("There are {} new task_ids to process.".format(
len(task_ids)))
pbar = tqdm(task_ids)
for t_id in pbar:
pbar.set_description("Processing task_id: {}".format(t_id))
try:
taskdoc = self._tasks.find_one({"task_id": dbid_to_int(t_id)})
m_id = self._match_material(taskdoc)
if not m_id:
m_id = self._create_new_material(taskdoc)
self._update_material(m_id, taskdoc)
except:
import traceback
logger.exception("<---")
logger.exception(
"There was an error processing task_id: {}".format(t_id))
logger.exception(traceback.format_exc())
logger.exception("--->")
logger.info("TasksMaterialsBuilder finished processing.")
def _create_new_material(self, taskdoc):
"""
Create a new material document.
Args:
taskdoc (dict): a JSON-like task document
Returns:
(int) - material_id of the new document
"""
doc = {"created_at": datetime.utcnow()}
doc["_tasksbuilder"] = {
"all_task_ids": [],
"prop_metadata": {
"labels": {},
"task_ids": {}
},
"updated_at": datetime.utcnow()
}
doc["spacegroup"] = taskdoc["output"]["spacegroup"]
doc["structure"] = taskdoc["output"]["structure"]
doc["material_id"] = dbid_to_str(
self._m_prefix,
self._counter.find_one_and_update(
{"_id": "materialid"}, {"$inc": {
"c": 1
}},
return_document=ReturnDocument.AFTER)["c"])
doc["sg_symbol"] = doc["spacegroup"]["symbol"]
doc["sg_number"] = doc["spacegroup"]["number"]
for x in [
"formula_anonymous", "formula_pretty", "formula_reduced_abc",
"elements", "nelements", "chemsys"
]:
doc[x] = taskdoc[x]
if "parent_structure" in taskdoc:
doc["parent_structure"] = taskdoc["parent_structure"]
t_struct = Structure.from_dict(
taskdoc["parent_structure"]["structure"])
doc["parent_structure"][
"formula_reduced_abc"] = t_struct.composition.reduced_formula
self._materials.insert_one(doc)
return doc["material_id"]
def run(self):
logger.info("MaterialsTaskBuilder starting...")
logger.info("Initializing list of all new task_ids to process ...")
previous_task_ids = []
for m in self._materials.find({}, {"_tasksbuilder.all_task_ids": 1}):
previous_task_ids.extend(m["_tasksbuilder"]["all_task_ids"])
q = {"state": "successful", "task_label": {"$in": self.supported_task_labels}}
if self.query:
common_keys = [k for k in q.keys() if k in self.query.keys()]
if common_keys:
raise ValueError("User query parameter cannot contain key(s): {}".
format(common_keys))
q.update(self.query)
all_task_ids = [dbid_to_str(self._t_prefix, t["task_id"]) for t in
self._tasks.find(q, {"task_id": 1})]
task_ids = [t_id for t_id in all_task_ids if t_id not in previous_task_ids]
logger.info("There are {} new task_ids to process.".format(len(task_ids)))
pbar = tqdm(task_ids)
for t_id in pbar:
pbar.set_description("Processing task_id: {}".format(t_id))
try:
taskdoc = self._tasks.find_one({"task_id": dbid_to_int(t_id)})
m_id = self._match_material(taskdoc)
if not m_id:
m_id = self._create_new_material(taskdoc)
self._update_material(m_id, taskdoc)
except:
import traceback
logger.exception("<---")
logger.exception("There was an error processing task_id: {}".format(t_id))
logger.exception(traceback.format_exc())
logger.exception("--->")
logger.info("TasksMaterialsBuilder finished processing.")
def _create_new_material(self, taskdoc):
"""
Create a new material document.
Args:
taskdoc (dict): a JSON-like task document
Returns:
(int) - material_id of the new document
"""
doc = {"created_at": datetime.utcnow()}
doc["_tasksbuilder"] = {"all_task_ids": [], "prop_metadata":
{"labels": {}, "task_ids": {}}, "updated_at": datetime.utcnow()}
doc["spacegroup"] = taskdoc["output"]["spacegroup"]
doc["structure"] = taskdoc["output"]["structure"]
doc["material_id"] = dbid_to_str(
self._m_prefix, self._counter.find_one_and_update(
{"_id": "materialid"}, {"$inc": {"c": 1}},
return_document=ReturnDocument.AFTER)["c"])
doc["sg_symbol"] = doc["spacegroup"]["symbol"]
doc["sg_number"] = doc["spacegroup"]["number"]
for x in ["formula_anonymous", "formula_pretty", "formula_reduced_abc", "elements",
"nelements", "chemsys"]:
doc[x] = taskdoc[x]
if "parent_structure" in taskdoc:
doc["parent_structure"] = taskdoc["parent_structure"]
t_struct = Structure.from_dict(taskdoc["parent_structure"]["structure"])
doc["parent_structure"]["formula_reduced_abc"] = t_struct.composition.reduced_formula
self._materials.insert_one(doc)
return doc["material_id"]
def _update_material(self, m_id, taskdoc):
"""
Update a material document based on a new task and using complex logic
Args:
m_id (int): material_id for material document to update
taskdoc (dict): a JSON-like task document
"""
# For each materials property, figure out what kind of task the data is currently based on
# as defined by the task label. This is used to decide if the new taskdoc is a type of
# calculation that provides higher quality data for that property
prop_tlabels = self._materials.find_one(
{"material_id": m_id},
{"_tasksbuilder.prop_metadata.labels": 1
})["_tasksbuilder"]["prop_metadata"]["labels"]
task_label = taskdoc[
"task_label"] # task label of new doc that updates this material
# figure out what materials properties need to be updated based on new task
for x in self.property_settings:
for p in x["properties"]:
# check if this is a valid task for getting the property
if task_label in x["quality_scores"]:
# assert: this is a valid task for the property
# but is it the "best" task for that property (highest quality score)?
t_quality = x["quality_scores"][task_label]
m_quality = x["quality_scores"].get(
prop_tlabels.get(p, None), None)
# check if this task's quality is better than existing data
# 3 possibilities:
# i) materials property data not present, so this is best
# ii) task quality higher based on task label
# iii) task quality equal to materials; use lowest energy task
if not m_quality or t_quality > m_quality \
or (t_quality == m_quality
and taskdoc["output"]["energy_per_atom"] <
self._materials.find_one({"material_id": m_id}, {
"_tasksbuilder": 1})["_tasksbuilder"]["prop_metadata"][
"energies"][p]):
# this task has better quality data
# figure out where the property data lives in the materials doc and
# in the task doc
materials_key = "{}.{}".format(x["materials_key"], p) \
if x.get("materials_key") else p
tasks_key = "{}.{}".format(x["tasks_key"], p) \
if x.get("tasks_key") else p
# insert property data AND metadata about this task
self._materials.\
update_one({"material_id": m_id},
{"$set": {materials_key: get_mongolike(taskdoc, tasks_key),
"_tasksbuilder.prop_metadata.labels.{}".format(p): task_label,
"_tasksbuilder.prop_metadata.task_ids.{}".format(p): dbid_to_str(
self._t_prefix, taskdoc["task_id"]),
"_tasksbuilder.prop_metadata.energies.{}".format(p): taskdoc["output"]["energy_per_atom"],
"_tasksbuilder.updated_at": datetime.utcnow()}})
# copy property to document root if in properties_root
# i.e., intentionally duplicate some data to the root level
if p in self.properties_root:
self._materials.\
update_one({"material_id": m_id},
{"$set": {p: get_mongolike(taskdoc, tasks_key)}})
# update the database to reflect that this task_id was already processed
self._materials.update_one({"material_id": m_id}, {
"$push": {
"_tasksbuilder.all_task_ids":
dbid_to_str(self._t_prefix, taskdoc["task_id"])
}
})
def _update_material(self, m_id, taskdoc):
"""
Update a material document based on a new task and using complex logic
Args:
m_id (int): material_id for material document to update
taskdoc (dict): a JSON-like task document
"""
# For each materials property, figure out what kind of task the data is currently based on
# as defined by the task label. This is used to decide if the new taskdoc is a type of
# calculation that provides higher quality data for that property
prop_tlabels = self._materials.find_one(
{"material_id": m_id}, {"_tasksbuilder.prop_metadata.labels": 1})[
"_tasksbuilder"]["prop_metadata"]["labels"]
task_label = taskdoc["task_label"] # task label of new doc that updates this material
# figure out what materials properties need to be updated based on new task
for x in self.property_settings:
for p in x["properties"]:
# check if this is a valid task for getting the property
if task_label in x["quality_scores"]:
# assert: this is a valid task for the property
# but is it the "best" task for that property (highest quality score)?
t_quality = x["quality_scores"][task_label]
m_quality = x["quality_scores"].get(prop_tlabels.get(p, None), None)
# check if this task's quality is better than existing data
# 3 possibilities:
# i) materials property data not present, so this is best
# ii) task quality higher based on task label
# iii) task quality equal to materials; use lowest energy task
if not m_quality or t_quality > m_quality \
or (t_quality == m_quality
and taskdoc["output"]["energy_per_atom"] <
self._materials.find_one({"material_id": m_id}, {
"_tasksbuilder": 1})["_tasksbuilder"]["prop_metadata"][
"energies"][p]):
# this task has better quality data
# figure out where the property data lives in the materials doc and
# in the task doc
materials_key = "{}.{}".format(x["materials_key"], p) \
if x.get("materials_key") else p
tasks_key = "{}.{}".format(x["tasks_key"], p) \
if x.get("tasks_key") else p
# insert property data AND metadata about this task
self._materials.\
update_one({"material_id": m_id},
{"$set": {materials_key: get_mongolike(taskdoc, tasks_key),
"_tasksbuilder.prop_metadata.labels.{}".format(p): task_label,
"_tasksbuilder.prop_metadata.task_ids.{}".format(p): dbid_to_str(
self._t_prefix, taskdoc["task_id"]),
"_tasksbuilder.prop_metadata.energies.{}".format(p): taskdoc["output"]["energy_per_atom"],
"_tasksbuilder.updated_at": datetime.utcnow()}})
# copy property to document root if in properties_root
# i.e., intentionally duplicate some data to the root level
if p in self.properties_root:
self._materials.\
update_one({"material_id": m_id},
{"$set": {p: get_mongolike(taskdoc, tasks_key)}})
# update the database to reflect that this task_id was already processed
self._materials.update_one({"material_id": m_id},
{"$push": {"_tasksbuilder.all_task_ids": dbid_to_str(
self._t_prefix, taskdoc["task_id"])}})
