File size: 5,520 Bytes
f002566
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
import torch
import torch.nn as nn
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.decomposition import PCA
import os
import sys
import json

# Import model
sys.path.append(os.path.join(os.path.dirname(__file__), '..', 'src'))
from model import TaxonomyAwareESM

RANK_NAMES = {
    0: "Superkingdom",
    1: "Phylum",
    2: "Class",
    3: "Order",
    4: "Family",
    5: "Genus",
    6: "Species"
}

TARGET_SPECIES = {
    "Homo sapiens": 9606,
    "Pongo abelii": 9601,
    "Hylobates muelleri": 9588
}

def load_species_vectors(vector_path):
    vectors = {}
    print(f"Loading species vectors from {vector_path}...")
    with open(vector_path, 'r') as f:
        for line in f:
            parts = line.strip().split('\t')
            if len(parts) >= 2:
                tax_id = int(parts[0])
                vec = json.loads(parts[1])
                vectors[tax_id] = vec
    return vectors

from adjustText import adjust_text

def visualize_embeddings(model_path, vector_path, output_dir):
    print("Visualizing Taxonomy Embeddings using PCA...")
    os.makedirs(output_dir, exist_ok=True)
    
    # 1. Load Model Weights
    try:
        checkpoint = torch.load(model_path, map_location='cpu')
        state_dict = checkpoint['model_state_dict'] if 'model_state_dict' in checkpoint else checkpoint
    except Exception as e:
        print(f"Error loading model: {e}")
        return

    # 2. Load Species Vectors (to find indices for targets)
    try:
        species_vectors = load_species_vectors(vector_path)
    except FileNotFoundError:
        print(f"Error: Vector file not found at {vector_path}")
        return
    
    # Verify targets exist
    target_indices = {} # Name -> [idx_0, idx_1, ... idx_6]
    for name, tax_id in TARGET_SPECIES.items():
        if tax_id in species_vectors:
            target_indices[name] = species_vectors[tax_id]
        else:
            print(f"Warning: TaxID {tax_id} ({name}) not found in species vectors.")

    # 3. Visualize per Rank
    for rank_idx in range(7):
        key = f"tax_embeddings.{rank_idx}.weight"
        if key not in state_dict:
            print(f"Missing weight for rank {rank_idx}")
            continue
            
        weight = state_dict[key].numpy()
        
        # PCA
        pca = PCA(n_components=2)
        transformed = pca.fit_transform(weight)
        
        # Plot
        plt.figure(figsize=(15, 15)) # Increased size
        
        # Background: All points (light grey)
        plt.scatter(transformed[:, 0], transformed[:, 1], c='lightgrey', alpha=0.3, s=20, label='Others') # lighter alpha, slightly larger s
        
        # Highlight Targets with Manual Positioning
        colors = ['red', 'blue', 'green']
        
        # Manual offsets/alignments for the 3 specific targets in order:
        # 1. Homo sapiens -> Top Right
        # 2. Pongo abelii -> Top Left
        # 3. Hylobates muelleri -> Bottom Right
        manual_positions = [
            {'xytext': (20, 20), 'ha': 'left', 'va': 'bottom'},  # Top-Right
            {'xytext': (-20, 20), 'ha': 'right', 'va': 'bottom'}, # Top-Left
            {'xytext': (20, -20), 'ha': 'left', 'va': 'top'}     # Bottom-Right
        ]

        for i, (name, indices) in enumerate(target_indices.items()):
            # indices is [v0, v1, ... v6]
            vocab_idx = indices[rank_idx]
            
            if vocab_idx < len(transformed):
                x, y = transformed[vocab_idx]
                
                # Marker
                plt.scatter(x, y, c=colors[i % len(colors)], s=300, edgecolor='black', zorder=10, marker='*')
                
                # Annotation (Manual Position)
                pos = manual_positions[i % len(manual_positions)]
                plt.annotate(name, 
                             xy=(x, y), 
                             xytext=pos['xytext'],
                             textcoords='offset points',
                             ha=pos['ha'], 
                             va=pos['va'],
                             fontsize=16, 
                             fontweight='bold', 
                             color='black',
                             arrowprops=dict(arrowstyle='->', color='black', lw=1.5))
            else:
                print(f"Index {vocab_idx} out of bounds for rank {rank_idx}")

        plt.title(f"Rank {rank_idx}: {RANK_NAMES.get(rank_idx, 'Unknown')} Embedding Space", fontsize=20)
        plt.xlabel("PC1", fontsize=14)
        plt.ylabel("PC2", fontsize=14)
        plt.tick_params(axis='both', which='major', labelsize=12)
        plt.grid(True, alpha=0.3)
        
        # No automatic adjust_text needed
        
        out_file = os.path.join(output_dir, f"rank_{rank_idx}_{RANK_NAMES[rank_idx]}_pca.png")
        plt.savefig(out_file, dpi=300, bbox_inches='tight') # High DPI
        plt.close()
        print(f"Saved plot to {out_file}")

if __name__ == "__main__":
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument("--model_path", type=str, required=True)
    parser.add_argument("--vector_path", type=str, default="data/species_vectors.tsv")
    parser.add_argument("--output_dir", type=str, default="outputs")
    args = parser.parse_args()
    
    visualize_embeddings(args.model_path, args.vector_path, args.output_dir)