Update app.py

app.py

@@ -1,52 +1,83 @@
+import json
 import os
 import pickle
 import re
-import json
 
 import gradio as gr
 import matplotlib.pyplot as plt
 import networkx as nx
 from tqdm import tqdm
 
+
 def load_json_from_path(path):
  with open(path, "r", encoding="utf8") as f:
  obj = json.loads(f.read())
 
  return obj
 
+
 class Visualizer:
 
  def __init__(self, cache_root="."):
+ self.iso_codes_to_names = load_json_from_path(os.path.join(cache_root, "iso_to_fullname.json"))
+ for code in self.iso_codes_to_names:
+ self.iso_codes_to_names[code] = re.sub("\(.*?\)", "", self.iso_codes_to_names[code])
+
  tree_lookup_path = os.path.join(cache_root, "lang_1_to_lang_2_to_tree_dist.json")
- self.tree_dist = load_json_from_path(tree_lookup_path)
+ tree_dist = load_json_from_path(tree_lookup_path)
+ distances = list()
+ for lang_1 in tree_dist:
+ if lang_1 not in self.iso_codes_to_names:
+ continue
+ for lang_2 in tree_dist[lang_1]:
+ if lang_2 not in self.iso_codes_to_names:
+ continue
+ if lang_1 != lang_2:
+ distances.append((self.iso_codes_to_names[lang_1], self.iso_codes_to_names[lang_2], tree_dist[lang_1][lang_2]))
+ min_dist = min(d for _, _, d in distances)
+ max_dist = max(d for _, _, d in distances)
+ self.tree_distances = [(entity1, entity2, (d - min_dist) / (max_dist - min_dist)) for entity1, entity2, d in distances]
 
  map_lookup_path = os.path.join(cache_root, "lang_1_to_lang_2_to_map_dist.json")
- self.map_dist = load_json_from_path(map_lookup_path)
- largest_value_map_dist = 0.0
- for _, values in self.map_dist.items():
- for _, value in values.items():
- largest_value_map_dist = max(largest_value_map_dist, value)
- for key1 in self.map_dist:
- for key2 in self.map_dist[key1]:
- self.map_dist[key1][key2] = self.map_dist[key1][key2] / largest_value_map_dist
+ map_dist = load_json_from_path(map_lookup_path)
+ distances = list()
+ for lang_1 in map_dist:
+ if lang_1 not in self.iso_codes_to_names:
+ continue
+ for lang_2 in map_dist[lang_1]:
+ if lang_2 not in self.iso_codes_to_names:
+ continue
+ if lang_1 != lang_2:
+ distances.append((self.iso_codes_to_names[lang_1], self.iso_codes_to_names[lang_2], map_dist[lang_1][lang_2]))
+ min_dist = min(d for _, _, d in distances)
+ max_dist = max(d for _, _, d in distances)
+ self.map_distances = [(entity1, entity2, (d - min_dist) / (max_dist - min_dist)) for entity1, entity2, d in distances]
 
  asp_dict_path = os.path.join(cache_root, "asp_dict.pkl")
  with open(asp_dict_path, 'rb') as dictfile:
  asp_sim = pickle.load(dictfile)
  lang_list = list(asp_sim.keys())
- self.asp_dist = dict()
+ asp_dist = dict()
  seen_langs = set()
  for lang_1 in lang_list:
  if lang_1 not in seen_langs:
  seen_langs.add(lang_1)
- self.asp_dist[lang_1] = dict()
+ asp_dist[lang_1] = dict()
  for index, lang_2 in enumerate(lang_list):
  if lang_2 not in seen_langs: # it's symmetric
- self.asp_dist[lang_1][lang_2] = 1 - asp_sim[lang_1][index]
-
- self.iso_codes_to_names = load_json_from_path(os.path.join(cache_root, "iso_to_fullname.json"))
- for code in self.iso_codes_to_names:
- self.iso_codes_to_names[code] = re.sub("\(.*?\)", "", self.iso_codes_to_names[code])
+ asp_dist[lang_1][lang_2] = 1 - asp_sim[lang_1][index]
+ distances = list()
+ for lang_1 in asp_dist:
+ if lang_1 not in self.iso_codes_to_names:
+ continue
+ for lang_2 in asp_dist[lang_1]:
+ if lang_2 not in self.iso_codes_to_names:
+ continue
+ if lang_1 != lang_2:
+ distances.append((self.iso_codes_to_names[lang_1], self.iso_codes_to_names[lang_2], asp_dist[lang_1][lang_2]))
+ min_dist = min(d for _, _, d in distances)
+ max_dist = max(d for _, _, d in distances)
+ self.asp_distances = [(entity1, entity2, (d - min_dist) / (max_dist - min_dist)) for entity1, entity2, d in distances]
 
  def visualize(self, distance_type, neighbor, num_neighbors):
  plt.clf()
@@ -56,46 +87,30 @@ class Visualizer:
  "Distance to the Lowest Common Ancestor in the Language Family Tree",
  "Angular Distance between the Frequencies of Phonemes"]
  if distance_type == "Distance to the Lowest Common Ancestor in the Language Family Tree":
- distance_measure = self.tree_dist
+ normalized_distances = self.tree_distances
  elif distance_type == "Angular Distance between the Frequencies of Phonemes":
- distance_measure = self.asp_dist
+ normalized_distances = self.asp_distances
  elif distance_type == "Physical Distance between Language Centroids on the Globe":
- distance_measure = self.map_dist
-
- distances = list()
-
- for lang_1 in distance_measure:
- if lang_1 not in self.iso_codes_to_names:
- continue
- for lang_2 in distance_measure[lang_1]:
- if lang_2 not in self.iso_codes_to_names:
- continue
- distances.append((self.iso_codes_to_names[lang_1], self.iso_codes_to_names[lang_2], distance_measure[lang_1][lang_2]))
+ normalized_distances = self.map_distances
 
  G = nx.Graph()
- min_dist = min(d for _, _, d in distances)
- max_dist = max(d for _, _, d in distances)
- normalized_distances = [(entity1, entity2, (d - min_dist) / (max_dist - min_dist)) for entity1, entity2, d in distances]
-
  d_dist = list()
  for entity1, entity2, d in tqdm(normalized_distances):
  if neighbor == entity2 or neighbor == entity1:
- if entity1 != entity2:
- d_dist.append(d)
+ d_dist.append(d)
  thresh = sorted(d_dist)[num_neighbors]
  neighbors = set()
  for entity1, entity2, d in tqdm(normalized_distances):
- if d < thresh and (neighbor == entity2 or neighbor == entity1) and (entity1 != entity2):
+ if d <= thresh and (neighbor == entity2 or neighbor == entity1) and len(neighbors) < num_neighbors + 1:
  neighbors.add(entity1)
  neighbors.add(entity2)
- spring_tension = (thresh - d) * 10 # for vis purposes
+ spring_tension = (thresh - d) * 100 # for vis purposes
  G.add_edge(entity1, entity2, weight=spring_tension)
  neighbors.remove(neighbor)
  for entity1, entity2, d in tqdm(normalized_distances):
  if entity2 in neighbors and entity1 in neighbors:
- if entity1 != entity2:
- spring_tension = thresh - d
- G.add_edge(entity1, entity2, weight=spring_tension)
+ spring_tension = thresh - d
+ G.add_edge(entity1, entity2, weight=spring_tension)
 
  pos = nx.spring_layout(G, weight="weight") # Positions for all nodes
  edges = G.edges(data=True)
@@ -106,7 +121,7 @@ class Visualizer:
  # nx.draw_networkx_edges(G, pos, edgelist=edges_not_connected_to_specific_node, edge_color='gray', alpha=0.1, width=1)
  for u, v, d in edges:
  if u == neighbor or v == neighbor:
- nx.draw_networkx_edge_labels(G, pos, edge_labels={(u, v): round((thresh - (d['weight'] / 10)) * 10, 2)}, font_color="red", alpha=0.4) # reverse modifications
+ nx.draw_networkx_edge_labels(G, pos, edge_labels={(u, v): round((thresh - (d['weight'] / 100)) * 10, 2)}, font_color="red", alpha=0.4) # reverse modifications
  nx.draw_networkx_labels(G, pos, font_size=14, font_family='sans-serif', font_color='green')
  nx.draw_networkx_labels(G, pos, labels={neighbor: neighbor}, font_size=14, font_family='sans-serif', font_color='red')
  plt.title(f'Graph of {distance_type}')
@@ -116,7 +131,7 @@ class Visualizer:
 
 
 if __name__ == '__main__':
- vis = Visualizer(cache_root=".")
+ vis = Visualizer(cache_root="Preprocessing/multilinguality")
  text_selection = [f"{vis.iso_codes_to_names[iso_code]}" for iso_code in vis.iso_codes_to_names]
  iface = gr.Interface(fn=vis.visualize,
  inputs=[gr.Dropdown(["Physical Distance between Language Centroids on the Globe",