Created
February 19, 2024 17:18
-
-
Save botcs/9d290eea4738aea30b2eb98cb32b9191 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| import torch | |
| class MicroCluster: | |
| def __init__(self, x, label, max_radius=0.1): | |
| self.N = 1 | |
| self.LS = x.clone() | |
| self.SS = x.pow(2) | |
| self.label = label | |
| self.max_radius = max_radius | |
| def add_point(self, x): | |
| self.N += 1 | |
| self.LS += x | |
| self.SS += x.pow(2) | |
| @property | |
| def centroid(self): | |
| return self.LS / self.N | |
| @property | |
| def radius(self): | |
| return torch.sqrt(self.SS / self.N - self.centroid.pow(2)).mean() | |
| def can_accept(self, x): | |
| return torch.norm(x - self.centroid) <= self.max_radius | |
| def merge(self, mc): | |
| self.N += mc.N | |
| self.LS += mc.LS | |
| self.SS += mc.SS | |
| class MClassification: | |
| def __init__(self, max_radius=0.1, max_clusters=100): | |
| self.max_radius = max_radius | |
| self.max_clusters = max_clusters | |
| self.micro_clusters = [] | |
| def fit_initial(self, X, labels): | |
| for x, label in zip(X, labels): | |
| self.micro_clusters.append(MicroCluster(x, label, self.max_radius)) | |
| def find_farthest_mc_from_point_with_label(self, x, label): | |
| farthest_distance = -1 | |
| farthest_index = None | |
| for i, mc in enumerate(self.micro_clusters): | |
| if mc.label == label: | |
| distance = torch.norm(x - mc.centroid) | |
| if distance > farthest_distance: | |
| farthest_distance = distance | |
| farthest_index = i | |
| return farthest_index | |
| def find_closest_mc_to_mc(self, mc_index): | |
| target_mc = self.micro_clusters[mc_index] | |
| closest_distance = float('inf') | |
| closest_index = None | |
| for i, mc in enumerate(self.micro_clusters): | |
| if i != mc_index: | |
| distance = torch.norm(target_mc.centroid - mc.centroid) | |
| if distance < closest_distance: | |
| closest_distance = distance | |
| closest_index = i | |
| return closest_index | |
| def merge_clusters(self, index1, index2): | |
| self.micro_clusters[index1].merge(self.micro_clusters[index2]) | |
| del self.micro_clusters[index2] | |
| def predict_and_update(self, x): | |
| if not self.micro_clusters: | |
| return -1 # Placeholder for no clusters case | |
| distances = [torch.norm(x - mc.centroid) for mc in self.micro_clusters] | |
| nearest_mc_index = distances.index(min(distances)) | |
| nearest_mc = self.micro_clusters[nearest_mc_index] | |
| if nearest_mc.can_accept(x): | |
| nearest_mc.add_point(x) | |
| else: | |
| # When a new point doesn't fit in existing clusters | |
| # Step 1: Find the farthest MC with the same label | |
| farthest_mc_index = self.find_farthest_mc_from_point_with_label(x, nearest_mc.label) | |
| # Step 2: Find the closest MC to the farthest MC | |
| if farthest_mc_index is not None: | |
| closest_to_farthest_mc_index = self.find_closest_mc_to_mc(farthest_mc_index) | |
| # Step 3: Merge these two MCs | |
| if closest_to_farthest_mc_index is not None: | |
| self.merge_clusters(farthest_mc_index, closest_to_farthest_mc_index) | |
| # Add the new point as a new MC | |
| self.micro_clusters.append(MicroCluster(x, nearest_mc.label, self.max_radius)) | |
| return nearest_mc.label |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment