def test_merge_with():
dicts = {1: 1, 2: 2}, {1: 10, 2: 20}
assert merge_with(sum, *dicts) == {1: 11, 2: 22}
assert merge_with(tuple, *dicts) == {1: (1, 10), 2: (2, 20)}
dicts = {1: 1, 2: 2, 3: 3}, {1: 10, 2: 20}
assert merge_with(sum, *dicts) == {1: 11, 2: 22, 3: 3}
assert merge_with(tuple, *dicts) == {1: (1, 10), 2: (2, 20), 3: (3,)}
assert not merge_with(sum)
def test_merge_with():
dicts = {1: 1, 2: 2}, {1: 10, 2: 20}
assert merge_with(sum, *dicts) == {1: 11, 2: 22}
assert merge_with(tuple, *dicts) == {1: (1, 10), 2: (2, 20)}
dicts = {1: 1, 2: 2, 3: 3}, {1: 10, 2: 20}
assert merge_with(sum, *dicts) == {1: 11, 2: 22, 3: 3}
assert merge_with(tuple, *dicts) == {1: (1, 10), 2: (2, 20), 3: (3,)}
assert not merge_with(sum)
def test_merge_with(self):
D, kw = self.D, self.kw
dicts = D({1: 1, 2: 2}), D({1: 10, 2: 20})
assert merge_with(sum, *dicts, **kw) == D({1: 11, 2: 22})
assert merge_with(tuple, *dicts, **kw) == D({1: (1, 10), 2: (2, 20)})
dicts = D({1: 1, 2: 2, 3: 3}), D({1: 10, 2: 20})
assert merge_with(sum, *dicts, **kw) == D({1: 11, 2: 22, 3: 3})
assert merge_with(tuple, *dicts, **kw) == D({1: (1, 10), 2: (2, 20), 3: (3,)})
assert not merge_with(sum)
def test_merge_with(self):
D, kw = self.D, self.kw
dicts = D({1: 1, 2: 2}), D({1: 10, 2: 20})
assert merge_with(sum, *dicts, **kw) == D({1: 11, 2: 22})
assert merge_with(tuple, *dicts, **kw) == D({1: (1, 10), 2: (2, 20)})
dicts = D({1: 1, 2: 2, 3: 3}), D({1: 10, 2: 20})
assert merge_with(sum, *dicts, **kw) == D({1: 11, 2: 22, 3: 3})
assert merge_with(tuple, *dicts, **kw) == D({1: (1, 10), 2: (2, 20), 3: (3,)})
assert not merge_with(sum)
def compute_dev_s_scores():
## TODO: Write the values to the dataframe dynamically instead of at the very end
key_name = 'dev_s'
model_names, task_names, ngrams_list, seeds_list = get_settings_list()
args = get_arguments()
args.inp_dir = eutils.mkdir_p(f'./out_dir/{key_name}')
avg_scores = []
for outer_idx, (task, model, ngram) in enumerate(itertools.product(task_names, model_names, ngrams_list)):
print(f'\nKey_name: {key_name}, Model: {model}, task: {task}, ngram: {ngram}\n')
args.task_name = task
args.model = model
args.ngram = ngram
tmp_res = []
for inner_idx, seed in enumerate(seeds_list):
args.seed = seed ## Set the seed
tmp_res.append(get_model_scores(args))
dict_of_list = merge_with(list, *tmp_res)
dict_of_avg_val = {key: statistics.mean(val) for key, val in dict_of_list.items()}
avg_scores.append({'model': model, 'task_name': task, 'ngram': ngram, 'word_order_sensitivity':word_order_sensitivity(dict_of_avg_val['accuracy']),
**dict_of_avg_val})
df = pd.DataFrame(avg_scores)
file_name = args.out_dir / f'{key_name}_scores.csv'
df.to_csv(file_name, index=False)
def deep_merge(*ds):
def combine(vals):
if len(vals) == 1 or not all(isinstance(v, dict) for v in vals):
return vals[-1]
else:
return deep_merge(*vals)
return merge_with(combine, *ds)
def trials_to_dimensions(trials, precision=6, remove_single_values=True):
"""
Converts hyperparameter configurations into the Plotly dimensions, e.g. used
to plot parallel coordinates.
Args:
trials (list): Hyperparameter configurations for a sequence of trials.
A trial is represented by a dict of hyperparameters and corresponding values.
precision (int, optional): The precision for displaying values of type float.
remove_single_values (bool, optional): If true, remove hyperparameters with a
single unique value.
Returns:
:class:`list` of :class:`dict`: A list of dictionaries describing each
hyperparameter dimension.
"""
# TODO: If floats are below 1., switch to exponential notation
# TODO: Scale axis to prevent ticks from overlapping
def to_dimension_dict(hparam):
label, values = hparam
elem = values[0]
if isinstance(elem, str):
unique = list(sorted(set(values)))
stoi = {v: i for i, v in enumerate(unique)}
tick_values = list(range(len(unique)))
tick_text = unique
values = [stoi[v] for v in values]
elif isinstance(elem, float):
fmt = '{{:.{}g}}'
fmt = fmt.format(precision)
tick_values = list(sorted(set(values)))
tick_text = [fmt.format(v) for v in tick_values]
else:
tick_values = list(sorted(set(values)))
tick_text = [str(v) for v in tick_values]
dct = dict(range=[min(values), max(values)],
label=label,
values=values,
tickvals=tick_values,
ticktext=tick_text)
return dct
hparams = dicttoolz.merge_with(list, trials)
dimensions = [to_dimension_dict(hparam) for hparam in hparams.items()]
if remove_single_values:
dimensions = list(filter(lambda d: len(d['tickvals']) > 1, dimensions))
return dimensions
def step(self, actions):
actions = np.split(actions, self.env_nums)
result = [
env.step(np.squeeze(action))
for env, action in zip(self.envs, actions)
]
obs, rews, dones, infos = zip(*result)
self._obs = np.stack(obs)
infos = merge_with(np.array, *infos)
return self._obs, np.stack(rews)[:, np.newaxis], \
np.stack(dones)[:, np.newaxis], infos
def _deepmerge(dicts):
# merge_with expects a non-variadic function
for maybe_a_dict in reversed(dicts):
if not isinstance(maybe_a_dict, dict):
# If we've got any non-dicts, the last non-dict wins.
return maybe_a_dict
else:
# Otherwise we want to merge all these dicts, using deepmerge on any
# collisions
return merge_with(_deepmerge, *dicts)
def find_corners(self) -> List[int]:
edges = {}
edges_dict_list = [{
min(edge, edge[::-1]): tile.id
for edge in tile.edges
} for tile in self.tiles]
edges = merge_with(lambda x: x, *edges_dict_list)
freqs = frequencies(
concat((value for value in edges.values() if len(value) == 2)))
corners = [id for id, count in freqs.items() if count == 2]
if len(corners) != 4:
raise ValueError("Wrong number of corners!")
return corners
def computeRuleHitsForFileSet(self, xliffs):
"""
For each file in the given filename -> PO object dictionary,
compute the Rule -> Hits dictonary.
Stores the information in the current instance.
Does not return anything
"""
# Compute dict with sorted & prettified filenames
self.files = sorted(xliffs.keys())
# Add all futures to the executor
futures = [
self.executor.submit(self.computeRuleHits, filename)
for filename in xliffs.keys()
]
# Process the results in first-received order. Also keep track of rule performance
self.fileRuleHits = collections.defaultdict(dict)
n_finished = 0
# Intermediate result storage
raw_results = collections.defaultdict(
dict) # filename -> {rule: result}
for future in concurrent.futures.as_completed(futures):
# Extract result
for filename, rule, result in future.result():
self.fileRuleHits[filename][rule] = result
# Track progress
n_finished += 1
if n_finished % 1000 == 0:
percent_finished = n_finished * 100. / len(futures)
print("Rule computation finished {0:.2f} %".format(
percent_finished))
# Compute total stats by file
self.statsByFile = {
filename:
merge(self.ruleHitsToSeverityCountMap(ruleHits),
{"translation_url": self.translationURLs[filename]})
for filename, ruleHits in self.fileRuleHits.items()
}
# Compute map filename -> {rule: numHits for rule}
self.statsByFileAndRule = {
filename: valmap(len, ruleHits)
for filename, ruleHits in self.fileRuleHits.items()
}
# Compute map rule -> numHits for rule
self.totalStatsByRule = merge_with(sum,
*(self.statsByFileAndRule.values()))
def step(self, actions):
actions = np.split(actions, self.proc_nums * self.env_nums_per_proc)
for index, parent_pipe in enumerate(self.parent_pipes):
parent_pipe.send((
'step',
actions[
index * self.env_nums_per_proc: (index +1) * self.env_nums_per_proc
]
))
results = []
for parent_pipe in self.parent_pipes:
results += parent_pipe.recv()
obs, rews, dones, infos = zip(*results)
self._obs = np.stack(obs)
infos = merge_with(np.array, *infos)
return self._obs, np.stack(rews)[:, np.newaxis], \
np.stack(dones)[:, np.newaxis], infos
def test_merge_with_non_dict_mappings():
class Foo(Mapping):
def __init__(self, d):
self.d = d
def __iter__(self):
return iter(self.d)
def __getitem__(self, key):
return self.d[key]
def __len__(self):
return len(self.d)
d = Foo({1: 1})
assert merge(d) is d or merge(d) == {1: 1}
assert merge_with(sum, d) == {1: 1}
def computeRuleHitsForFileSet(self, poFiles):
"""
For each file in the given filename -> PO object dictionary,
compute the Rule -> Hits dictonary.
Stores the information in the current instance.
Does not return anything
"""
# Compute dict with sorted & prettified filenames
self.files = sorted(poFiles.keys())
# Add all futures to the executor
futures = list(itertools.chain(*(self.computeRuleHits(po, filename)
for filename, po in poFiles.items())))
# Process the results in first-received order. Also keep track of rule performance
self.fileRuleHits = collections.defaultdict(dict)
n_finished = 0
# Intermediate result storage
raw_results = collections.defaultdict(dict) # filename -> {rule: result}
for future in concurrent.futures.as_completed(futures):
# Extract result
filename, rule, result = future.result()
self.fileRuleHits[filename][rule] = result
# Track progress
n_finished += 1
if n_finished % 1000 == 0:
percent_finished = n_finished * 100. / len(futures)
print("Rule computation finished {0:.2f} %".format(percent_finished))
# Compute total stats by file
self.statsByFile = {
filename: merge(self.ruleHitsToSeverityCountMap(ruleHits), {
"translation_url": self.translationURLs[filename]})
for filename, ruleHits in self.fileRuleHits.items()
}
# Compute map filename -> {rule: numHits for rule}
self.statsByFileAndRule = {
filename: valmap(len, ruleHits)
for filename, ruleHits in self.fileRuleHits.items()
}
# Compute map rule -> numHits for rule
self.totalStatsByRule = merge_with(sum, *(self.statsByFileAndRule.values()))
import sys
from toolz.dicttoolz import merge_with, valfilter
def parse(line):
parts = line.split()
return (parts[1], parts[7])
pairs = [parse(line) for line in open('1')]
d = merge_with(lambda x: [z for y in x for z in y], *map(lambda x: dict([(x[0], [x[1]])]), pairs))
print(d)
avail = ''.join(list(d.keys()))
print(avail)
for v in d.values():
for c in v:
avail = avail.replace(c, '')
avail = sorted(avail)
order = ''
current = ''
def filter_avail(s):
out = []
for x in s:
pres = list(valfilter(lambda y: x in y, d).keys())
good = True
for p in pres:
if p not in order:
good = False
break
if good:
def test_merge_with_iterable_arg():
dicts = {1: 1, 2: 2}, {1: 10, 2: 20}
assert merge_with(sum, *dicts) == {1: 11, 2: 22}
assert merge_with(sum, dicts) == {1: 11, 2: 22}
assert merge_with(sum, iter(dicts)) == {1: 11, 2: 22}
def test_merge_with_iterable_arg(self):
D, kw = self.D, self.kw
dicts = D({1: 1, 2: 2}), D({1: 10, 2: 20})
assert merge_with(sum, *dicts, **kw) == D({1: 11, 2: 22})
assert merge_with(sum, dicts, **kw) == D({1: 11, 2: 22})
assert merge_with(sum, iter(dicts), **kw) == D({1: 11, 2: 22})
def GetRecommendedEvents(self, request, context):
session = DBSession()
try:
user_id = request.user_id
k_events = request.k_events
user_recommendation_score = session.query(UserEvent)\
.filter(UserEvent.user_id == user_id, EventRecommendation.score.isnot(None))\
.join(EventRecommendation, UserEvent.event_id == EventRecommendation.event_id)\
.with_entities(EventRecommendation.score)
dictionary = {}
for item in user_recommendation_score:
dictionary = merge_with(sum, dictionary, item[0])
event_ids = list(dictionary.keys())
# check durations
duration_query = session.query(EventDuration).filter(EventDuration.event_id.in_(event_ids))\
.group_by(EventDuration.event_id)\
.with_entities(EventDuration.event_id, func.min(EventDuration.start))
event_ids_to_be_removed = []
for item in duration_query:
if datetime.now().replace(tzinfo=utc) > item[1].replace(
tzinfo=utc):
event_ids_to_be_removed.append(str(item[0]))
event_ids_future = list(
set(event_ids).difference(set(event_ids_to_be_removed)))
# get score and convert to prob using softmax
events_score = [dictionary[x] for x in event_ids_future]
if len(events_score) == 0:
return personalization_service.GetRecommendedEventsResponse(
event_collection=[])
prob = softmax(events_score)
# events to recommended
size = min(k_events, len(event_ids_future))
choice = np.random.choice(event_ids_future,
p=prob,
size=size,
replace=False)
query_events = session.query(Event).filter(
Event.id.in_(choice)).all()
data = map(
lambda event: common.Event(
id=event.id,
organization_id=event.organization_id,
location_id=getInt32Value(event.location_id),
description=event.description,
name=event.name,
cover_image_url=getStringValue(event.cover_image_url),
cover_image_hash=getStringValue(event.cover_image_hash),
poster_image_url=getStringValue(event.poster_image_url),
poster_image_hash=getStringValue(event.poster_image_hash),
profile_image_url=getStringValue(event.profile_image_url),
profile_image_hash=getStringValue(event.profile_image_hash
),
attendee_limit=event.attendee_limit,
contact=getStringValue(event.contact),
registration_due_date=getTimeStamp(event.
registration_due_date),
),
query_events,
)
return personalization_service.GetRecommendedEventsResponse(
event_collection=data)
except Exception as e:
session.rollback()
raise Exception(f"Something went wrong: {e}")
finally:
session.close()
def test_merge_with_iterable_arg():
dicts = {1: 1, 2: 2}, {1: 10, 2: 20}
assert merge_with(sum, *dicts) == {1: 11, 2: 22}
assert merge_with(sum, dicts) == {1: 11, 2: 22}
assert merge_with(sum, iter(dicts)) == {1: 11, 2: 22}
def test_merge_with_iterable_arg(self):
D, kw = self.D, self.kw
dicts = D({1: 1, 2: 2}), D({1: 10, 2: 20})
assert merge_with(sum, *dicts, **kw) == D({1: 11, 2: 22})
assert merge_with(sum, dicts, **kw) == D({1: 11, 2: 22})
assert merge_with(sum, iter(dicts), **kw) == D({1: 11, 2: 22})
line = line.split()
line = [stem(i) for i in line]
for s in line:
d[s] += 1
return {k: d[k] for k in d.keys() if not drop_word(k)}
if __name__ == "__main__":
workflow = (
glob,
mapcat(open),
mapcat(str.split),
map(stem),
frequencies,
keyfilter(drop_word),
)
wordcount = compose(*reversed(workflow))
billboard = wordcount("lyrics/billboard/*")
dylan = wordcount("lyrics/dylan/*")
m = merge_with(sum, normalise(dylan), normalise(billboard, sign=-1))
print("\nDylan:")
col_print(sorted(m, key=m.get)[-50:])
print("\nBillboard:")
col_print(sorted(m, key=m.get)[:50])
