fiftyone.utils.random

Generates a random sample of size k from the given collection such that the expected histogram of path values in the sample is uniform.

Example:

import fiftyone as fo
import fiftyone.utils.random as four
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("cifar10", split="train")

# Sample proportional to label length
weights = dataset.values(F("ground_truth.label").strlen())
view1 = four.weighted_sample(dataset, 10000, weights)

# Now take a balanced sample from this unbalanced sample
view2 = four.balanced_sample(view1, 2000, "ground_truth.label")

# Plot results
plot1 = fo.CategoricalHistogram("ground_truth.label", init_view=dataset)
plot2 = fo.CategoricalHistogram(
    "ground_truth.label",
    order=lambda kv: -len(kv[0]),  # order by label length
    init_view=view1,
)
plot3 = fo.CategoricalHistogram("ground_truth.label", init_view=view2)
plot = fo.ViewGrid([plot1, plot2, plot3])
plot.show()

Generates a random partition of the samples in the collection according to the specified split fractions.

Example:

import fiftyone as fo
import fiftyone.utils.random as four
import fiftyone.zoo as foz

# A dataset with `ground_truth` detections and no tags
dataset = (
    foz.load_zoo_dataset("quickstart")
    .select_fields("ground_truth")
    .set_field("tags", [])
).clone()

#
# Generate a random sample and encode results via tags
#

four.random_split(dataset, {"train": 0.7, "test": 0.2, "val": 0.1})

print(dataset.count_sample_tags())
# {'train': 140, 'test': 40, 'val': 20}

#
# Generate a random sample in-memory
#

view1, view2 = four.random_split(dataset, [0.5, 0.5])

assert len(view1) + len(view2) == len(dataset)
assert set(view1.values("id")).isdisjoint(set(view2.values("id")))

Generates a random sample of size k from the given collection such that the probability of selecting each sample is proportional to the given per-sample weights.

Example:

import fiftyone as fo
import fiftyone.utils.random as four
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("cifar10", split="train")

# Sample proportional to label length
weights = dataset.values(F("ground_truth.label").strlen())
sample = four.weighted_sample(dataset, 10000, weights)

# Plot results
plot = fo.CategoricalHistogram(
    "ground_truth.label",
    order=lambda kv: -len(kv[0]),  # order by label length
    init_view=sample,
)
plot.show()