def filter_flats_list(config, constraint_name, flats_list, fetch_details=True, past_flats=None):
"""
Filter the available flats list. Then, filter it according to criteria.
:param config: A config dict.
:param constraint_name: The constraint name that the ``flats_list`` should
satisfy.
:param flats_list: The initial list of flat objects to filter.
:param fetch_details: Whether additional details should be fetched between
the two passes.
:param past_flats: The list of already fetched flats
:return: A dict mapping flat status and list of flat objects.
"""
# Add the flatisfy metadata entry and prepare the flat objects
flats_list = metadata.init(flats_list, constraint_name)
# Get the associated constraint from config
try:
constraint = config["constraints"][constraint_name]
except KeyError:
LOGGER.error(
"Missing constraint %s. Skipping filtering for these posts.",
constraint_name,
)
return {"new": [], "duplicate": [], "ignored": []}
first_pass_result = collections.defaultdict(list)
second_pass_result = collections.defaultdict(list)
third_pass_result = collections.defaultdict(list)
# Do a first pass with the available infos to try to remove as much
# unwanted postings as possible
if config["passes"] > 0:
first_pass_result = flatisfy.filters.first_pass(flats_list, constraint, config)
else:
first_pass_result["new"] = flats_list
# Load additional infos
if fetch_details:
past_ids = {x["id"]: x for x in past_flats} if past_flats else {}
for i, flat in enumerate(first_pass_result["new"]):
details = None
use_cache = past_ids.get(flat["id"])
if use_cache:
LOGGER.debug("Skipping details download for %s.", flat["id"])
details = use_cache
else:
if flat["id"].split("@")[1] in ["seloger", "leboncoin"]:
try:
details = fetch.fetch_details_rate_limited(config, flat["id"])
except RateLimitException:
time.sleep(60)
details = fetch.fetch_details_rate_limited(config, flat["id"])
else:
details = fetch.fetch_details(config, flat["id"])
first_pass_result["new"][i] = tools.merge_dicts(flat, details)
# Do a second pass to consolidate all the infos we found and make use of
# additional infos
if config["passes"] > 1:
second_pass_result = flatisfy.filters.second_pass(first_pass_result["new"], constraint, config)
else:
second_pass_result["new"] = first_pass_result["new"]
# Do a third pass to deduplicate better
if config["passes"] > 2:
third_pass_result = flatisfy.filters.third_pass(second_pass_result["new"], config)
else:
third_pass_result["new"] = second_pass_result["new"]
return {
"new": third_pass_result["new"],
"duplicate": (
first_pass_result["duplicate"] + second_pass_result["duplicate"] + third_pass_result["duplicate"]
),
"ignored": (first_pass_result["ignored"] + second_pass_result["ignored"] + third_pass_result["ignored"]),
}
def filter_flats_list(config, constraint_name, flats_list, fetch_details=True):
"""
Filter the available flats list. Then, filter it according to criteria.
:param config: A config dict.
:param constraint_name: The constraint name that the ``flats_list`` should
satisfy.
:param fetch_details: Whether additional details should be fetched between
the two passes.
:param flats_list: The initial list of flat objects to filter.
:return: A dict mapping flat status and list of flat objects.
"""
# Add the flatisfy metadata entry and prepare the flat objects
flats_list = metadata.init(flats_list, constraint_name)
# Get the associated constraint from config
try:
constraint = config["constraints"][constraint_name]
except KeyError:
LOGGER.error(
"Missing constraint %s. Skipping filtering for these posts.",
constraint_name)
return {"new": [], "duplicate": [], "ignored": []}
first_pass_result = collections.defaultdict(list)
second_pass_result = collections.defaultdict(list)
third_pass_result = collections.defaultdict(list)
# Do a first pass with the available infos to try to remove as much
# unwanted postings as possible
if config["passes"] > 0:
first_pass_result = flatisfy.filters.first_pass(
flats_list, constraint, config)
else:
first_pass_result["new"] = flats_list
# Load additional infos
if fetch_details:
for i, flat in enumerate(first_pass_result["new"]):
details = fetch.fetch_details(config, flat["id"])
first_pass_result["new"][i] = tools.merge_dicts(flat, details)
# Do a second pass to consolidate all the infos we found and make use of
# additional infos
if config["passes"] > 1:
second_pass_result = flatisfy.filters.second_pass(
first_pass_result["new"], constraint, config)
else:
second_pass_result["new"] = first_pass_result["new"]
# Do a third pass to deduplicate better
if config["passes"] > 2:
third_pass_result = flatisfy.filters.third_pass(
second_pass_result["new"], config)
else:
third_pass_result["new"] = second_pass_result["new"]
return {
"new":
third_pass_result["new"],
"duplicate":
(first_pass_result["duplicate"] + second_pass_result["duplicate"] +
third_pass_result["duplicate"]),
"ignored":
(first_pass_result["ignored"] + second_pass_result["ignored"] +
third_pass_result["ignored"])
}
def deep_detect(flats_list):
"""
Deeper detection of duplicates based on any available data.
:param flats_list: A list of flats dicts.
:return: A tuple of the deduplicated list of flat dicts and the list of all
the flats objects that should be removed and considered as duplicates (they
were already merged).
"""
photo_cache = ImageCache()
LOGGER.info("Running deep duplicates detection.")
matching_flats = collections.defaultdict(list)
for i, flat1 in enumerate(flats_list):
matching_flats[flat1["id"]].append(flat1["id"])
for j, flat2 in enumerate(flats_list):
if i <= j:
continue
if flat2["id"] in matching_flats[flat1["id"]]:
continue
n_common_items = 0
try:
# They should have the same area, up to one unit
assert abs(flat1["area"] - flat2["area"]) < 1
n_common_items += 1
# They should be at the same price, up to one unit
assert abs(flat1["cost"] - flat2["cost"]) < 1
n_common_items += 1
# They should have the same number of bedrooms if this was
# fetched for both
if flat1["bedrooms"] and flat2["bedrooms"]:
assert flat1["bedrooms"] == flat2["bedrooms"]
n_common_items += 1
# They should have the same utilities (included or excluded for
# both of them), if this was fetched for both
if flat1["utilities"] and flat2["utilities"]:
assert flat1["utilities"] == flat2["utilities"]
n_common_items += 1
# They should have the same number of rooms if it was fetched
# for both of them
if flat1["rooms"] and flat2["rooms"]:
assert flat1["rooms"] == flat2["rooms"]
n_common_items += 1
# They should have the same postal code, if available
if (flat1["flatisfy"].get("postal_code", None)
and flat2["flatisfy"].get("postal_code", None)):
assert (flat1["flatisfy"]["postal_code"] ==
flat2["flatisfy"]["postal_code"])
n_common_items += 1
# TODO: Compare texts (one is included in another? fuzzymatch?)
# They should have the same phone number if it was fetched for
# both
flat1_phone = homogeneize_phone_number(flat1["phone"])
flat2_phone = homogeneize_phone_number(flat2["phone"])
if flat1_phone and flat2_phone:
assert flat1_phone == flat2_phone
n_common_items += 10 # Counts much more that the rest
# They should have at least one photo in common if there
# are some photos
if flat1["photos"] and flat2["photos"]:
n_common_photos = find_number_common_photos(
photo_cache, flat1["photos"], flat2["photos"])
assert n_common_photos > 1
min_number_photos = min(len(flat1["photos"]),
len(flat2["photos"]))
# Either all the photos are the same, or there are at least
# three common photos.
if n_common_photos == min_number_photos:
n_common_items += 15
else:
n_common_items += 5 * min(n_common_photos, 3)
# Minimal score to consider they are duplicates
assert n_common_items >= 15
except (AssertionError, TypeError):
# Skip and consider as not duplicates whenever the conditions
# are not met
# TypeError occurs when an area or a cost is None, which should
# not be considered as duplicates
continue
# Mark flats as duplicates
LOGGER.info(("Found duplicates using deep detection: (%s, %s). "
"Score is %d."), flat1["id"], flat2["id"],
n_common_items)
matching_flats[flat1["id"]].append(flat2["id"])
matching_flats[flat2["id"]].append(flat1["id"])
if photo_cache.total():
LOGGER.debug("Photo cache: hits: %d%% / misses: %d%%.",
photo_cache.hit_rate(), photo_cache.miss_rate())
seen_ids = []
duplicate_flats = []
unique_flats_list = []
for flat_id in [flat["id"] for flat in flats_list]:
if flat_id in seen_ids:
continue
seen_ids.extend(matching_flats[flat_id])
to_merge = sorted([
flat
for flat in flats_list if flat["id"] in matching_flats[flat_id]
],
key=lambda flat:
next(i for
(i, backend) in enumerate(BACKENDS_PRECEDENCE)
if flat["id"].endswith(backend)),
reverse=True)
unique_flats_list.append(tools.merge_dicts(*to_merge))
# The ID of the added merged flat will be the one of the last item
# in ``matching_flats``. Then, any flat object that was before in
# the ``matching_flats`` list is to be considered as a duplicate
# and should have a ``duplicate`` status.
duplicate_flats.extend(to_merge[:-1])
return unique_flats_list, duplicate_flats
def detect(flats_list, key="id", merge=True, should_intersect=False):
"""
Detect obvious duplicates within a given list of flats.
There may be duplicates found, as some queries could overlap (especially
since when asking for a given place, websites tend to return housings in
nearby locations as well). We need to handle them, by either deleting the
duplicates (``merge=False``) or merging them together in a single flat
object.
:param flats_list: A list of flats dicts.
:param key: The flat dicts key on which the duplicate detection should be
done.
:param merge: Whether the found duplicates should be merged or we should
only keep one of them.
:param should_intersect: Set to ``True`` if the values in the flat dicts
are lists and you want to deduplicate on non-empty intersection (typically
if they have a common url).
:return: A tuple of the deduplicated list of flat dicts and the list of all
the flats objects that should be removed and considered as duplicates (they
were already merged).
"""
# ``seen`` is a dict mapping aggregating the flats by the deduplication
# keys. We basically make buckets of flats for every key value. Flats in
# the same bucket should be merged together afterwards.
seen = collections.defaultdict(list)
for flat in flats_list:
if should_intersect:
# We add each value separately. We will add some flats multiple
# times, but we deduplicate again on id below to compensate.
for value in flat.get(key, []):
seen[value].append(flat)
else:
seen[flat.get(key, None)].append(flat)
# Generate the unique flats list based on these buckets
unique_flats_list = []
# Keep track of all the flats that were removed by deduplication
duplicate_flats = []
for flat_key, matching_flats in seen.items():
if flat_key is None:
# If the key is None, it means Weboob could not load the data. In
# this case, we consider every matching item as being independant
# of the others, to avoid over-deduplication.
unique_flats_list.extend(matching_flats)
else:
# Sort matching flats by backend precedence
matching_flats.sort(key=lambda flat: next(
i for (i, backend) in enumerate(BACKENDS_PRECEDENCE)
if flat["id"].endswith(backend)),
reverse=True)
if len(matching_flats) > 1:
LOGGER.info("Found duplicates using key \"%s\": %s.", key,
[flat["id"] for flat in matching_flats])
# Otherwise, check the policy
if merge:
# If a merge is requested, do the merge
unique_flats_list.append(tools.merge_dicts(*matching_flats))
else:
# Otherwise, just keep the most important of them
unique_flats_list.append(matching_flats[-1])
# The ID of the added merged flat will be the one of the last item
# in ``matching_flats``. Then, any flat object that was before in
# the ``matching_flats`` list is to be considered as a duplicate
# and should have a ``duplicate`` status.
duplicate_flats.extend(matching_flats[:-1])
if should_intersect:
# We added some flats twice with the above method, let's deduplicate on
# id.
unique_flats_list, _ = detect(unique_flats_list,
key="id",
merge=True,
should_intersect=False)
return unique_flats_list, duplicate_flats
def deep_detect(flats_list, config):
"""
Deeper detection of duplicates based on any available data.
:param flats_list: A list of flats dicts.
:param config: A config dict.
:return: A tuple of the deduplicated list of flat dicts and the list of all
the flats objects that should be removed and considered as duplicates
(they were already merged).
"""
if config["serve_images_locally"]:
storage_dir = os.path.join(config["data_directory"], "images")
else:
storage_dir = None
photo_cache = ImageCache(storage_dir=storage_dir)
LOGGER.info("Running deep duplicates detection.")
matching_flats = collections.defaultdict(list)
for i, flat1 in enumerate(flats_list):
matching_flats[flat1["id"]].append(flat1["id"])
for j, flat2 in enumerate(flats_list):
if i <= j:
continue
if flat2["id"] in matching_flats[flat1["id"]]:
continue
n_common_items = get_duplicate_score(
flat1, flat2, photo_cache,
config["duplicate_image_hash_threshold"])
# Minimal score to consider they are duplicates
if n_common_items >= config["duplicate_threshold"]:
# Mark flats as duplicates
LOGGER.info(
("Found duplicates using deep detection: (%s, %s). Score is %d."
),
flat1["id"],
flat2["id"],
n_common_items,
)
matching_flats[flat1["id"]].append(flat2["id"])
matching_flats[flat2["id"]].append(flat1["id"])
if photo_cache.total():
LOGGER.debug(
"Photo cache: hits: %d%% / misses: %d%%.",
photo_cache.hit_rate(),
photo_cache.miss_rate(),
)
seen_ids = []
duplicate_flats = []
unique_flats_list = []
for flat_id in [flat["id"] for flat in flats_list]:
if flat_id in seen_ids:
continue
seen_ids.extend(matching_flats[flat_id])
to_merge = sorted(
[
flat
for flat in flats_list if flat["id"] in matching_flats[flat_id]
],
key=lambda flat: next(i for (i, backend) in enumerate(
BACKENDS_BY_PRECEDENCE) if flat["id"].endswith(backend)),
reverse=True,
)
unique_flats_list.append(tools.merge_dicts(*to_merge))
# The ID of the added merged flat will be the one of the last item
# in ``matching_flats``. Then, any flat object that was before in
# the ``matching_flats`` list is to be considered as a duplicate
# and should have a ``duplicate`` status.
duplicate_flats.extend(to_merge[:-1])
return unique_flats_list, duplicate_flats
