Source code for cleanlab.benchmarking.noise_generation

# Copyright (C) 2017-2022  Cleanlab Inc.
# This file is part of cleanlab.
#
# cleanlab is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# cleanlab is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with cleanlab.  If not, see <https://www.gnu.org/licenses/>.


"""
Helper methods that are useful for benchmarking cleanlab’s core algorithms.
These methods introduce synthetic noise into the labels of a classification dataset.
Specifically, this module provides methods for generating valid noise matrices (for which learning with noise is possible),
generating noisy labels given a noise matrix, generating valid noise matrices with a specific trace value, and more.
"""

import numpy as np
from cleanlab.internal.util import value_counts
import warnings


[docs]def noise_matrix_is_valid(noise_matrix, py, *, verbose=False): """Given a prior `py` representing ``p(true_label=k)``, checks if the given `noise_matrix` is a learnable matrix. Learnability means that it is possible to achieve better than random performance, on average, for the amount of noise in `noise_matrix`. Parameters ---------- noise_matrix : np.array An array of shape ``(K, K)`` representing the conditional probability matrix ``P(label=k_s|true_label=k_y)`` containing the fraction of examples in every class, labeled as every other class. Assumes columns of `noise_matrix` sum to 1. py : np.array An array of shape ``(K,)`` representing the fraction (prior probability) of each true class label, ``P(true_label = k)``. Returns ------- bool Whether the noise matrix is a learnable matrix. """ # Number of classes K = len(py) # let's assume some number of training examples for code readability, # but it doesn't matter what we choose as it's not actually used. N = float(10000) ps = np.dot(noise_matrix, py) # P(true_label=k) # P(label=k, true_label=k') joint_noise = np.multiply(noise_matrix, py) # / float(N) # Check that joint_probs is valid probability matrix if not (abs(joint_noise.sum() - 1.0) < 1e-6): return False # Check that noise_matrix is a valid matrix # i.e. check p(label=k)*p(true_label=k) < p(label=k, true_label=k) for i in range(K): C = N * joint_noise[i][i] E1 = N * joint_noise[i].sum() - C E2 = N * joint_noise.T[i].sum() - C O = N - E1 - E2 - C if verbose: print( "E1E2/C", round(E1 * E2 / C), "E1", round(E1), "E2", round(E2), "C", round(C), "|", round(E1 * E2 / C + E1 + E2 + C), "|", round(E1 * E2 / C), "<", round(O), ) print( round(ps[i] * py[i]), "<", round(joint_noise[i][i]), ":", ps[i] * py[i] < joint_noise[i][i], ) if not (ps[i] * py[i] < joint_noise[i][i]): return False return True
[docs]def generate_noisy_labels(true_labels, noise_matrix): """Generates noisy `labels` from perfect labels `true_labels`, "exactly" yielding the provided `noise_matrix` between `labels` and `true_labels`. Below we provide a for loop implementation of what this function does. We do not use this implementation as it is not a fast algorithm, but it explains as Python pseudocode what is happening in this function. Parameters ---------- true_labels : np.array An array of shape ``(N,)`` representing perfect labels, without any noise. Contains K distinct natural number classes, 0, 1, ..., K-1. noise_matrix : np.array An array of shape ``(K, K)`` representing the conditional probability matrix ``P(label=k_s|true_label=k_y)`` containing the fraction of examples in every class, labeled as every other class. Assumes columns of `noise_matrix` sum to 1. Returns ------- labels : np.array An array of shape ``(N,)`` of noisy labels. Examples -------- .. code:: python # Generate labels count_joint = (noise_matrix * py * len(y)).round().astype(int) labels = np.array(y) for k_s in range(K): for k_y in range(K): if k_s != k_y: idx_flip = np.where((labels==k_y)&(true_label==k_y))[0] if len(idx_flip): # pragma: no cover labels[np.random.choice( idx_flip, count_joint[k_s][k_y], replace=False, )] = k_s """ # Make y a numpy array, if it is not true_labels = np.asarray(true_labels) # Number of classes K = len(noise_matrix) # Compute p(true_label=k) py = value_counts(true_labels) / float(len(true_labels)) # Counts of pairs (labels, y) count_joint = (noise_matrix * py * len(true_labels)).astype(int) # Remove diagonal entries as they do not involve flipping of labels. np.fill_diagonal(count_joint, 0) # Generate labels labels = np.array(true_labels) for k in range(K): # Iterate over true_label == k # Get the noisy labels that have non-zero counts labels_per_class = np.where(count_joint[:, k] != 0)[0] # Find out how many of each noisy label we need to flip to label_counts = count_joint[labels_per_class, k] # Create a list of the new noisy labels noise = [labels_per_class[i] for i, c in enumerate(label_counts) for z in range(c)] # Randomly choose y labels for class k and set them to the noisy labels. idx_flip = np.where((labels == k) & (true_labels == k))[0] if len(idx_flip) and len(noise) and len(idx_flip) >= len(noise): # pragma: no cover labels[np.random.choice(idx_flip, len(noise), replace=False)] = noise # Validate that labels indeed produces the correct noise_matrix (or close to it) # Compute the actual noise matrix induced by labels # counts = confusion_matrix(labels, true_labels).astype(float) # new_noise_matrix = counts / counts.sum(axis=0) # assert(np.linalg.norm(noise_matrix - new_noise_matrix) <= 2) return labels
[docs]def generate_noise_matrix_from_trace( K, trace, *, max_trace_prob=1.0, min_trace_prob=1e-5, max_noise_rate=1 - 1e-5, min_noise_rate=0.0, valid_noise_matrix=True, py=None, frac_zero_noise_rates=0.0, seed=0, max_iter=10000, ): """Generates a ``K x K`` noise matrix ``P(label=k_s|true_label=k_y)`` with ``np.sum(np.diagonal(noise_matrix))`` equal to the given `trace`. Parameters ---------- K : int Creates a noise matrix of shape ``(K, K)``. Implies there are K classes for learning with noisy labels. trace : float Sum of diagonal entries of array of random probabilities returned. max_trace_prob : float Maximum probability of any entry in the trace of the return matrix. min_trace_prob : float Minimum probability of any entry in the trace of the return matrix. max_noise_rate : float Maximum noise_rate (non-diagonal entry) in the returned np.array. min_noise_rate : float Minimum noise_rate (non-diagonal entry) in the returned np.array. valid_noise_matrix : bool, default=True If ``True``, returns a matrix having all necessary conditions for learning with noisy labels. In particular, ``p(true_label=k)p(label=k) < p(true_label=k,label=k)`` is satisfied. This requires that ``trace > 1``. py : np.array An array of shape ``(K,)`` representing the fraction (prior probability) of each true class label, ``P(true_label = k)``. This argument is **required** when ``valid_noise_matrix=True``. frac_zero_noise_rates : float The fraction of the ``n*(n-1)`` noise rates that will be set to 0. Note that if you set a high trace, it may be impossible to also have a low fraction of zero noise rates without forcing all non-1 diagonal values. Instead, when this happens we only guarantee to produce a noise matrix with `frac_zero_noise_rates` *or higher*. The opposite occurs with a small trace. seed : int Seeds the random number generator for numpy. max_iter : int, default=10000 The max number of tries to produce a valid matrix before returning ``None``. Returns ------- noise_matrix : np.array or None An array of shape ``(K, K)`` representing the noise matrix ``P(label=k_s|true_label=k_y)`` with `trace` equal to ``np.sum(np.diagonal(noise_matrix))``. This a conditional probability matrix and a left stochastic matrix. Returns ``None`` if `max_iter` is exceeded. """ if valid_noise_matrix and trace <= 1: raise ValueError( "trace = {}. trace > 1 is necessary for a".format(trace) + " valid noise matrix to be returned (valid_noise_matrix == True)" ) if valid_noise_matrix and py is None and K > 2: raise ValueError( "py must be provided (not None) if the input parameter" + " valid_noise_matrix == True" ) if K <= 1: raise ValueError("K must be >= 2, but K = {}.".format(K)) if max_iter < 1: return None np.random.seed(seed) # Special (highly constrained) case with faster solution. # Every 2 x 2 noise matrix with trace > 1 is valid because p(y) is not used if K == 2: if frac_zero_noise_rates >= 0.5: # Include a single zero noise rate noise_mat = np.array( [ [1.0, 1 - (trace - 1.0)], [0.0, trace - 1.0], ] ) return noise_mat if np.random.rand() > 0.5 else np.rot90(noise_mat, k=2) else: # No zero noise rates diag = generate_n_rand_probabilities_that_sum_to_m(2, trace) noise_matrix = np.array( [ [diag[0], 1 - diag[1]], [1 - diag[0], diag[1]], ] ) return noise_matrix # K > 2 for z in range(max_iter): noise_matrix = np.zeros(shape=(K, K)) # Randomly generate noise_matrix diagonal. nm_diagonal = generate_n_rand_probabilities_that_sum_to_m( n=K, m=trace, max_prob=max_trace_prob, min_prob=min_trace_prob, ) np.fill_diagonal(noise_matrix, nm_diagonal) # Randomly distribute number of zero-noise-rates across columns num_col_with_noise = K - np.count_nonzero(1 == nm_diagonal) num_zero_noise_rates = int(K * (K - 1) * frac_zero_noise_rates) # Remove zeros already in [1,0,..,0] columns num_zero_noise_rates -= (K - num_col_with_noise) * (K - 1) num_zero_noise_rates = np.maximum(num_zero_noise_rates, 0) # Prevent negative num_zero_noise_rates_per_col = ( randomly_distribute_N_balls_into_K_bins( N=num_zero_noise_rates, K=num_col_with_noise, max_balls_per_bin=K - 2, # 2 = one for diagonal, and one to sum to 1 min_balls_per_bin=0, ) if K > 2 else np.array([0, 0]) ) # Special case when K == 2 stack_nonzero_noise_rates_per_col = list(K - 1 - num_zero_noise_rates_per_col)[::-1] # Randomly generate noise rates for columns with noise. for col in np.arange(K)[nm_diagonal != 1]: num_noise = stack_nonzero_noise_rates_per_col.pop() # Generate num_noise noise_rates for the given column. noise_rates_col = list( generate_n_rand_probabilities_that_sum_to_m( n=num_noise, m=1 - nm_diagonal[col], max_prob=max_noise_rate, min_prob=min_noise_rate, ) ) # Randomly select which rows of the noisy column to assign the # random noise rates rows = np.random.choice( [row for row in range(K) if row != col], num_noise, replace=False ) for row in rows: noise_matrix[row][col] = noise_rates_col.pop() if not valid_noise_matrix or noise_matrix_is_valid(noise_matrix, py): return noise_matrix return None
[docs]def generate_n_rand_probabilities_that_sum_to_m( n, m, *, max_prob=1.0, min_prob=0.0, ): """ Generates `n` random probabilities that sum to `m`. When ``min_prob=0`` and ``max_prob = 1.0``, use ``np.random.dirichlet(np.ones(n))*m`` instead. Parameters ---------- n : int Length of array of random probabilities to be returned. m : float Sum of array of random probabilities that is returned. max_prob : float, default=1.0 Maximum probability of any entry in the returned array. Must be between 0 and 1. min_prob : float, default=0.0 Minimum probability of any entry in the returned array. Must be between 0 and 1. """ epsilon = 1e-6 # Imprecision allowed for inequalities with floats if n == 0: return np.array([]) if (max_prob + epsilon) < m / float(n): raise ValueError( "max_prob must be greater or equal to m / n, but " + "max_prob = " + str(max_prob) + ", m = " + str(m) + ", n = " + str(n) + ", m / n = " + str(m / float(n)) ) if min_prob > (m + epsilon) / float(n): raise ValueError( "min_prob must be less or equal to m / n, but " + "max_prob = " + str(max_prob) + ", m = " + str(m) + ", n = " + str(n) + ", m / n = " + str(m / float(n)) ) # When max_prob = 1, min_prob = 0, the next two lines are equivalent to: # intermediate = np.sort(np.append(np.random.uniform(0, 1, n-1), [0, 1])) # result = (intermediate[1:] - intermediate[:-1]) * m result = np.random.dirichlet(np.ones(n)) * m min_val = min(result) max_val = max(result) while max_val > (max_prob + epsilon): new_min = min_val + (max_val - max_prob) # This adjustment prevents the new max from always being max_prob. adjustment = (max_prob - new_min) * np.random.rand() result[np.argmin(result)] = new_min + adjustment result[np.argmax(result)] = max_prob - adjustment min_val = min(result) max_val = max(result) min_val = min(result) max_val = max(result) while min_val < (min_prob - epsilon): min_val = min(result) max_val = max(result) new_max = max_val - (min_prob - min_val) # This adjustment prevents the new min from always being min_prob. adjustment = (new_max - min_prob) * np.random.rand() result[np.argmax(result)] = new_max - adjustment result[np.argmin(result)] = min_prob + adjustment min_val = min(result) max_val = max(result) return result
[docs]def randomly_distribute_N_balls_into_K_bins( N, # int K, # int *, max_balls_per_bin=None, min_balls_per_bin=None, ): """Returns a uniformly random numpy integer array of length N that sums to K. Parameters ---------- N : int Number of balls. K : int Number of bins. max_balls_per_bin : int Ensure that each bin contains at most `max_balls_per_bin` balls. min_balls_per_bin : int Ensure that each bin contains at least `min_balls_per_bin` balls. Returns ------- np.array """ if N == 0: return np.zeros(K, dtype=int) if max_balls_per_bin is None: max_balls_per_bin = N else: max_balls_per_bin = min(max_balls_per_bin, N) if min_balls_per_bin is None: min_balls_per_bin = 0 else: min_balls_per_bin = min(min_balls_per_bin, N / K) if N / float(K) > max_balls_per_bin: N = max_balls_per_bin * K arr = np.round( generate_n_rand_probabilities_that_sum_to_m( n=K, m=1, max_prob=max_balls_per_bin / float(N), min_prob=min_balls_per_bin / float(N), ) * N ) while sum(arr) != N: while sum(arr) > N: # pragma: no cover arr[np.argmax(arr)] -= 1 while sum(arr) < N: arr[np.argmin(arr)] += 1 return arr.astype(int)