def run_mds(matrix: DistMatrix, max_iter: int, step_size: int, init_type: int,
embedding: np.ndarray, state: TaskState):
res = Result(embedding=embedding)
iterations_done = 0
init = embedding
state.set_status("Running...")
oldstress = np.finfo(np.float).max
while True:
step_iter = min(max_iter - iterations_done, step_size)
mds = MDS(
dissimilarity="precomputed", n_components=2,
n_init=1, max_iter=step_iter,
init_type=init_type, init_data=init
)
mdsfit = mds(matrix)
iterations_done += step_iter
embedding, stress = mdsfit.embedding_, mdsfit.stress_
emb_norm = np.sqrt(np.sum(embedding ** 2, axis=1)).sum()
if emb_norm > 0:
stress /= emb_norm
res.embedding = embedding
state.set_partial_result(res)
state.set_progress_value(100 * iterations_done / max_iter)
if iterations_done >= max_iter or stress == 0 or \
(oldstress - stress) < mds.params["eps"]:
return res
init = embedding
oldstress = stress
if state.is_interruption_requested():
return res
def _run(
data: Table,
group_by_attrs: List[Variable],
aggregations: Dict[Variable, Set[str]],
result: Result,
state: TaskState,
) -> Result:
def progress(part):
state.set_progress_value(part * 100)
if state.is_interruption_requested():
raise Exception
state.set_status("Aggregating")
# group table rows
if result.group_by is None:
result.group_by = data.groupby(group_by_attrs)
state.set_partial_result(result)
aggregations = {
var: [(agg, AGGREGATIONS[agg].function)
for agg in sorted(aggs, key=AGGREGATIONS_ORD.index)]
for var, aggs in aggregations.items()
}
result.result_table = result.group_by.aggregate(
aggregations, wrap_callback(progress, 0.2, 1))
return result
def run(corpus: Corpus, statistics: Tuple[int, str], state: TaskState) -> None:
"""
This function runs the computation for new features.
All results will be reported as a partial results.
Parameters
----------
corpus
The corpus on which the computation is held.
statistics
Tuple of statistic pairs to be computed:
(statistics id, string pattern)
state
State used to report progress and partial results.
"""
# callback is called for each corpus element statistics time
tick_values = iter(np.linspace(0, 100, len(corpus) * len(statistics)))
def advance():
state.set_progress_value(next(tick_values))
for s, patern in statistics:
fun = STATISTICS_FUNCTIONS[s]
result = fun(corpus, patern, advance)
if result is not None:
result = result + (ComputeValue(fun, patern),)
state.set_partial_result((s, patern, result))
def run_mds(matrix: DistMatrix, max_iter: int, step_size: int, init_type: int,
embedding: np.ndarray, state: TaskState):
res = Result(embedding=embedding)
iterations_done = 0
init = embedding
state.set_status("Running...")
oldstress = np.finfo(np.float).max
while True:
step_iter = min(max_iter - iterations_done, step_size)
mds = MDS(
dissimilarity="precomputed", n_components=2,
n_init=1, max_iter=step_iter,
init_type=init_type, init_data=init
)
mdsfit = mds(matrix)
iterations_done += step_iter
embedding, stress = mdsfit.embedding_, mdsfit.stress_
emb_norm = np.sqrt(np.sum(embedding ** 2, axis=1)).sum()
if emb_norm > 0:
stress /= emb_norm
res.embedding = embedding
state.set_partial_result(res)
state.set_progress_value(100 * iterations_done / max_iter)
if iterations_done >= max_iter or stress == 0 or \
(oldstress - stress) < mds.params["eps"]:
return res
init = embedding
oldstress = stress
if state.is_interruption_requested():
return res
def _prepare_dir_and_save_images(paths_queue, dir_name, target_size,
previously_saved, state: TaskState):
"""
This function prepares a directory structure and calls function
that saves images.
Parameters
----------
previously_saved : int
Number of saved images in the previous process. If the process is
resumed it is non-zero.
"""
res = Result(paths=paths_queue)
if previously_saved == 0:
_clean_dir(dir_name)
steps = len(paths_queue) + previously_saved
loader = ImageLoader()
while res.paths:
from_path, to_path = res.paths.popleft()
_save_an_image(loader, from_path, to_path, target_size)
state.set_progress_value((1 - len(res.paths) / steps) * 100)
state.set_partial_result(res)
if state.is_interruption_requested():
return res
return res
def run_embedding(
images: Table,
file_paths_attr: Variable,
embedder_name: str,
state: TaskState,
) -> Result:
"""
Run the embedding process
Parameters
----------
images
Data table with images to embed.
file_paths_attr
The column of the table with images.
embedder_name
The name of selected embedder.
state
State object used for controlling and progress.
Returns
-------
The object that holds embedded images, skipped images, and number
of skipped images.
"""
embedder = ImageEmbedder(model=embedder_name)
file_paths = images[:, file_paths_attr].metas.flatten()
file_paths_mask = file_paths == file_paths_attr.Unknown
file_paths_valid = file_paths[~file_paths_mask]
# init progress bar and fuction
ticks = iter(np.linspace(0.0, 100.0, file_paths_valid.size))
def advance(success=True):
if state.is_interruption_requested():
embedder.set_canceled()
if success:
state.set_progress_value(next(ticks))
try:
emb, skip, n_skip = embedder(images,
col=file_paths_attr,
callback=advance)
except EmbeddingConnectionError:
# recompute ticks to go from current state to 100
ticks = iter(np.linspace(next(ticks), 100.0, file_paths_valid.size))
state.set_partial_result("squeezenet")
embedder = ImageEmbedder(model="squeezenet")
emb, skip, n_skip = embedder(images,
col=file_paths_attr,
callback=advance)
return Result(embedding=emb, skip_images=skip, num_skipped=n_skip)
def run_vizrank(compute_score: Callable, iterate_states: Callable,
saved_state: Optional[Iterable], scores: List, progress: int,
state_count: int, task: TaskState):
task.set_status("Getting combinations...")
task.set_progress_value(0.1)
states = iterate_states(saved_state)
task.set_status("Getting scores...")
res = Result(queue=Queue(), scores=None)
scores = scores.copy()
can_set_partial_result = True
def do_work(st, next_st):
try:
score = compute_score(st)
if score is not None:
pos = bisect_left(scores, score)
res.queue.put_nowait(
QueuedScore(position=pos,
score=score,
state=st,
next_state=next_st))
scores.insert(pos, score)
except Exception: # ignore current state in case of any problem
pass
res.scores = scores.copy()
def reset_flag():
nonlocal can_set_partial_result
can_set_partial_result = True
state = None
next_state = next(states)
try:
while True:
if task.is_interruption_requested():
return res
task.set_progress_value(int(progress * 100 / max(1, state_count)))
progress += 1
state = copy.copy(next_state)
next_state = copy.copy(next(states))
do_work(state, next_state)
# for simple scores (e.g. correlations widget) and many feature
# combinations, the 'partial_result_ready' signal (emitted by
# invoking 'task.set_partial_result') was emitted too frequently
# for a longer period of time and therefore causing the widget
# being unresponsive
if can_set_partial_result:
task.set_partial_result(res)
can_set_partial_result = False
Timer(0.01, reset_flag).start()
except StopIteration:
do_work(state, None)
task.set_partial_result(res)
return res
def compute_secondary_clusters(embedding: Table, result: Result, state: TaskState):
if not result.clusters.groups or not embedding:
result.clusters.secondary_table = None
else:
state.set_status("Finding secondary clusters...")
hulls = {k: v[2] for k, v in result.clusters.groups.items()}
clusters = result.clusters.table
domain = clusters and clusters.domain["Clusters"]
table = cluster_additional_points(embedding, hulls, domain)
result.clusters.secondary_table = table
state.set_partial_result(("secondary_clusters", result))
def _run(
corpus: Corpus,
words: List[str],
scoring_methods: List[str],
aggregation: str,
additional_params: dict,
state: TaskState,
) -> None:
"""
Perform word scoring with selected scoring methods
Parameters
----------
corpus
Corpus of documents
words
List of words used for scoring
scoring_methods
Methods to score documents with
aggregation
Aggregation applied for each document on word scores
additional_params
Additional prameters for scores (e.g. embedding needs text language)
state
TaskState for reporting the task status and giving partial results
"""
def callback(i: float) -> None:
state.set_progress_value(i * 100)
if state.is_interruption_requested():
raise Exception
cb_part = 1 / (len(scoring_methods) + 1) # +1 for preprocessing
words = _preprocess_words(corpus, words,
wrap_callback(callback, end=cb_part))
if len(words) == 0:
raise Exception(
"Empty word list after preprocessing. Please provide a valid set of words."
)
for i, sm in enumerate(scoring_methods):
scoring_method = SCORING_METHODS[sm][1]
sig = signature(scoring_method)
add_params = {
k: v
for k, v in additional_params.items() if k in sig.parameters
}
scs = scoring_method(
corpus, words,
wrap_callback(callback,
start=(i + 1) * cb_part,
end=(i + 2) * cb_part), **add_params)
scs = AGGREGATIONS[aggregation](scs, axis=1)
state.set_partial_result((sm, aggregation, scs))
def compute_clusters(embedding: Table, result: Result, state: TaskState):
if not result.scores.table or not embedding:
result.clusters.table = None
result.clusters.groups = None
else:
state.set_status("Finding clusters...")
kwargs = {}
if result.clusters.epsilon is not None:
kwargs["eps"] = result.clusters.epsilon
clusters = annotate_projection(result.scores.table, embedding, **kwargs)
result.clusters.table = clusters[0]
result.clusters.groups = clusters[1]
result.clusters.epsilon = clusters[2]
state.set_partial_result(("clusters", result))
def run(data: Table, embedding: Optional[np.ndarray], state: TaskState):
res = Result(embedding=embedding)
# simulate wasteful calculation (increase 'steps')
step, steps = 0, 10
state.set_status("Calculating...")
while step < steps:
for _ in range(steps):
x_data = np.array(np.mean(data.X, axis=1))
if x_data.ndim == 2:
x_data = x_data.ravel()
y_data = np.random.rand(len(x_data))
embedding = np.vstack((x_data, y_data)).T
step += 1
if step % (steps / 10) == 0:
state.set_progress_value(100 * step / steps)
if state.is_interruption_requested():
return res
res.embedding = embedding
state.set_partial_result(res)
return res
def run(data: Table, embedding: Optional[np.ndarray], state: TaskState):
res = Result(embedding=embedding)
# simulate wasteful calculation (increase 'steps')
step, steps = 0, 10
state.set_status("Calculating...")
while step < steps:
for _ in range(steps):
x_data = np.array(np.mean(data.X, axis=1))
if x_data.ndim == 2:
x_data = x_data.ravel()
y_data = np.random.rand(len(x_data))
embedding = np.vstack((x_data, y_data)).T
step += 1
if step % (steps / 10) == 0:
state.set_progress_value(100 * step / steps)
if state.is_interruption_requested():
return res
res.embedding = embedding
state.set_partial_result(res)
return res
def run_freeviz(data: Table, projector: FreeViz, state: TaskState):
res = Result(projector=projector, projection=None)
step, steps = 0, MAX_ITERATIONS
initial = res.projector.components_.T
state.set_status("Calculating...")
while True:
# Needs a copy because projection should not be modified inplace.
# If it is modified inplace, the widget and the thread hold a
# reference to the same object. When the thread is interrupted it
# is still modifying the object, but the widget receives it
# (the modified object) with a delay.
res.projection = res.projector(data).copy()
anchors = res.projector.components_.T
res.projector.initial = anchors
state.set_partial_result(res)
if np.allclose(initial, anchors, rtol=1e-5, atol=1e-4):
return res
initial = anchors
step += 1
state.set_progress_value(100 * step / steps)
if state.is_interruption_requested():
return res
def run_freeviz(data: Table, projector: FreeViz, state: TaskState):
res = Result(projector=projector, projection=None)
step, steps = 0, MAX_ITERATIONS
initial = res.projector.components_.T
state.set_status("Calculating...")
while True:
# Needs a copy because projection should not be modified inplace.
# If it is modified inplace, the widget and the thread hold a
# reference to the same object. When the thread is interrupted it
# is still modifying the object, but the widget receives it
# (the modified object) with a delay.
res.projection = res.projector(data).copy()
anchors = res.projector.components_.T
res.projector.initial = anchors
state.set_partial_result(res)
if np.allclose(initial, anchors, rtol=1e-5, atol=1e-4):
return res
initial = anchors
step += 1
state.set_progress_value(100 * step / steps)
if state.is_interruption_requested():
return res
def compute_scores(
data: Table,
genes: Table,
p_threshold: float,
p_value_fun: str,
scoring: str,
start: float,
end: float,
result: Result,
state: TaskState,
):
if not data or not genes:
result.scores.z_vals = None
result.scores.annotations = None
result.scores.p_vals = None
result.scores.table = None
else:
state.set_status("Computing scores...")
weights = np.array([15, 75, 10]) * (end - start) / 100
if not result.scores.z_vals:
result.scores.z_vals = AnnotateSamplesMeta.mann_whitney_test(
data)
state.set_partial_result(("scores", result))
state.set_progress_value(weights[0])
if state.is_interruption_requested():
return
if not result.scores.annotations or not result.scores.p_vals:
annot, p_vals = AnnotateSamplesMeta.assign_annotations(
result.scores.z_vals,
genes,
data,
p_value_fun=p_value_fun,
scoring=scoring)
result.scores.annotations = annot
result.scores.p_vals = p_vals
state.set_partial_result(("scores", result))
state.set_progress_value(weights[1])
if state.is_interruption_requested():
return
result.scores.table = AnnotateSamplesMeta.filter_annotations(
result.scores.annotations,
result.scores.p_vals,
p_threshold=p_threshold)
state.set_partial_result(("scores", result))
def worker(self, state: TaskState):
while True:
state.set_partial_result(self.update_frame())
time.sleep(1 / 10)