Create app.py

app.py

@@ -0,0 +1,138 @@
+import os
+import pickle
+import re
+
+import gradio as gr
+import matplotlib.pyplot as plt
+import networkx as nx
+from tqdm import tqdm
+
+from Utility.utils import load_json_from_path
+
+
+class Visualizer:
+
+    def __init__(self, cache_root="."):
+        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)
+
+        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
+
+        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()
+        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()
+            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])
+
+    def visualize(self, distance_type, neighbor, num_neighbors):
+        plt.clf()
+        plt.figure(figsize=(12, 12))
+
+        assert distance_type in ["Physical Distance between Language Centroids on the Globe",
+                                 "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
+        elif distance_type == "Angular Distance between the Frequencies of Phonemes":
+            distance_measure = self.asp_dist
+        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]))
+
+        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)
+        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):
+                neighbors.add(entity1)
+                neighbors.add(entity2)
+                spring_tension = (thresh - d) * 10  # 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)
+
+        pos = nx.spring_layout(G, weight="weight")  # Positions for all nodes
+        edges = G.edges(data=True)
+        nx.draw_networkx_nodes(G, pos, node_size=1, alpha=0.01)
+        edges_connected_to_specific_node = [(u, v) for u, v in G.edges() if u == neighbor or v == neighbor]
+        nx.draw_networkx_edges(G, pos, edgelist=edges_connected_to_specific_node, edge_color='orange', alpha=0.4, width=3)
+        # edges_not_connected_to_specific_node = [(u, v) for u, v in G.edges() if u != neighbor and v != neighbor]
+        # 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_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}')
+        plt.subplots_adjust(left=0, right=1, top=0.9, bottom=0)
+        plt.tight_layout()
+        return plt.gcf()
+
+
+if __name__ == '__main__':
+    vis = Visualizer(cache_root=".")
+    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",
+                                              "Distance to the Lowest Common Ancestor in the Language Family Tree",
+                                              "Angular Distance between the Frequencies of Phonemes"],
+                                             type="value",
+                                             value='Physical Distance between Language Centroids on the Globe',
+                                             label="Select the Type of Distance"),
+                                 gr.Dropdown(text_selection,
+                                             type="value",
+                                             value="German",
+                                             label="Select the second Language (type on your keyboard to find it quickly)"),
+                                 gr.Slider(minimum=0, maximum=100, step=1,
+                                           value=12,
+                                           label="How many Nearest Neighbors should be displayed?")
+                                 ],
+                         outputs=[gr.Plot(label="", show_label=False, format="png", container=True)],
+                         description="<br><br> This demo allows you to find the nearest neighbors of a language from the ISO 639-3 list according to several distance measurement functions. "
+                                     "For more information, check out our paper: https://arxiv.org/abs/2406.06403 and our text-to-speech tool, in which we make use of "
+                                     "this technique: https://github.com/DigitalPhonetics/IMS-Toucan <br><br>",
+                         fill_width=True,
+                         allow_flagging="never")
+    iface.launch()