File size: 9,667 Bytes
4fac556 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 | """
Export minimal dataset to binary format for fast client-side loading.
This creates a compact binary representation optimized for WebGL rendering.
"""
import struct
import json
import numpy as np
import pandas as pd
from pathlib import Path
import sys
import os
# Add parent directory to path
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from utils.data_loader import ModelDataLoader
from utils.dimensionality_reduction import DimensionReducer
from utils.embeddings import ModelEmbedder
def calculate_family_depths(df: pd.DataFrame) -> dict:
"""Calculate depth of each model in its family tree."""
depths = {}
def get_depth(model_id: str, visited: set = None) -> int:
if visited is None:
visited = set()
if model_id in visited:
return 0 # Cycle detected
visited.add(model_id)
if model_id in depths:
return depths[model_id]
parent_col = df.get('parent_model', pd.Series([None] * len(df), index=df.index))
model_row = df[df['model_id'] == model_id]
if model_row.empty:
depths[model_id] = 0
return 0
parent = model_row.iloc[0].get('parent_model')
if pd.isna(parent) or parent == '' or str(parent) == 'nan':
depths[model_id] = 0
return 0
parent_depth = get_depth(str(parent), visited.copy())
depth = parent_depth + 1
depths[model_id] = depth
return depth
for model_id in df['model_id'].unique():
if model_id not in depths:
get_depth(str(model_id))
return depths
def export_binary_dataset(df: pd.DataFrame, reduced_embeddings: np.ndarray, output_dir: Path):
"""
Export minimal dataset to binary format for fast client-side loading.
Binary format:
- Header (64 bytes): magic, version, counts, lookup table sizes
- Domain lookup table (32 bytes per domain)
- License lookup table (32 bytes per license)
- Family lookup table (32 bytes per family)
- Model records (16 bytes each): x, y, z, domain_id, license_id, family_id, flags
"""
output_dir.mkdir(parents=True, exist_ok=True)
print(f"Exporting {len(df)} models to binary format...")
# Ensure we have coordinates
if 'x' not in df.columns or 'y' not in df.columns:
if reduced_embeddings is None or len(reduced_embeddings) != len(df):
raise ValueError("Need reduced embeddings to generate coordinates")
df['x'] = reduced_embeddings[:, 0] if reduced_embeddings.shape[1] > 0 else 0.0
df['y'] = reduced_embeddings[:, 1] if reduced_embeddings.shape[1] > 1 else 0.0
df['z'] = reduced_embeddings[:, 2] if reduced_embeddings.shape[1] > 2 else 0.0
# Create lookup tables
# Domain = library_name
domains = sorted(df['library_name'].dropna().astype(str).unique())
domains = [d for d in domains if d and d != 'nan'][:255] # Limit to 255
# License
licenses = sorted(df['license'].dropna().astype(str).unique())
licenses = [l for l in licenses if l and l != 'nan'][:255] # Limit to 255
# Family ID mapping (use parent_model to create family groups)
family_depths = calculate_family_depths(df)
# Create family mapping: group models by root parent
def get_root_parent(model_id: str) -> str:
visited = set()
current = str(model_id)
while current in visited == False:
visited.add(current)
model_row = df[df['model_id'] == current]
if model_row.empty:
return current
parent = model_row.iloc[0].get('parent_model')
if pd.isna(parent) or parent == '' or str(parent) == 'nan':
return current
current = str(parent)
return current
root_parents = {}
family_counter = 0
for model_id in df['model_id'].unique():
root = get_root_parent(str(model_id))
if root not in root_parents:
root_parents[root] = family_counter
family_counter += 1
# Map each model to its family
model_to_family = {}
for model_id in df['model_id'].unique():
root = get_root_parent(str(model_id))
model_to_family[str(model_id)] = root_parents.get(root, 65535)
# Limit families to 65535 (u16 max)
if len(root_parents) > 65535:
# Use hash-based family IDs
import hashlib
for model_id in df['model_id'].unique():
root = get_root_parent(str(model_id))
family_hash = int(hashlib.md5(root.encode()).hexdigest()[:4], 16) % 65535
model_to_family[str(model_id)] = family_hash
# Prepare model records
records = []
model_ids = []
for idx, row in df.iterrows():
model_id = str(row['model_id'])
model_ids.append(model_id)
# Get coordinates
x = float(row.get('x', 0.0))
y = float(row.get('y', 0.0))
z = float(row.get('z', 0.0))
# Encode domain (library_name)
domain_str = str(row.get('library_name', ''))
domain_id = domains.index(domain_str) if domain_str in domains else 255
# Encode license
license_str = str(row.get('license', ''))
license_id = licenses.index(license_str) if license_str in licenses else 255
# Encode family
family_id = model_to_family.get(model_id, 65535)
# Encode flags
flags = 0
parent = row.get('parent_model')
if pd.isna(parent) or parent == '' or str(parent) == 'nan':
flags |= 0x01 # is_base_model
# Check if has children (simple check - could be improved)
children = df[df['parent_model'] == model_id]
if len(children) > 0:
flags |= 0x04 # has_children
elif not pd.isna(parent) and parent != '' and str(parent) != 'nan':
flags |= 0x02 # has_parent
# Pack record: f32 x, f32 y, f32 z, u8 domain, u8 license, u16 family, u8 flags
records.append(struct.pack('fffBBBH', x, y, z, domain_id, license_id, family_id, flags))
num_models = len(records)
# Write binary file
with open(output_dir / 'embeddings.bin', 'wb') as f:
# Header (64 bytes)
header = struct.pack('5sBIIIBBH50s',
b'HFVIZ', # magic (5 bytes)
1, # version (1 byte)
num_models, # num_models (4 bytes)
len(domains), # num_domains (4 bytes)
len(licenses), # num_licenses (4 bytes)
len(set(model_to_family.values())), # num_families (4 bytes)
0, # reserved (1 byte)
0, # reserved (1 byte)
0, # reserved (2 bytes)
b'\x00' * 50 # padding (50 bytes)
)
f.write(header)
# Domain lookup table (32 bytes per domain, null-terminated)
for domain in domains:
domain_bytes = domain.encode('utf-8')[:31]
f.write(domain_bytes.ljust(32, b'\x00'))
# License lookup table (32 bytes per license)
for license in licenses:
license_bytes = license.encode('utf-8')[:31]
f.write(license_bytes.ljust(32, b'\x00'))
# Model records
f.write(b''.join(records))
# Write model IDs JSON (separate file for string table)
with open(output_dir / 'model_ids.json', 'w') as f:
json.dump(model_ids, f)
# Write metadata JSON
metadata = {
'domains': domains,
'licenses': licenses,
'num_models': num_models,
'num_families': len(set(model_to_family.values())),
'version': 1
}
with open(output_dir / 'metadata.json', 'w') as f:
json.dump(metadata, f, indent=2)
binary_size = (output_dir / 'embeddings.bin').stat().st_size
json_size = (output_dir / 'model_ids.json').stat().st_size
print(f"✓ Exported {num_models} models")
print(f"✓ Binary size: {binary_size / 1024 / 1024:.2f} MB")
print(f"✓ Model IDs JSON: {json_size / 1024 / 1024:.2f} MB")
print(f"✓ Total: {(binary_size + json_size) / 1024 / 1024:.2f} MB")
print(f"✓ Domains: {len(domains)}")
print(f"✓ Licenses: {len(licenses)}")
print(f"✓ Families: {len(set(model_to_family.values()))}")
if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser(description='Export dataset to binary format')
parser.add_argument('--output', type=str, default='backend/cache/binary', help='Output directory')
parser.add_argument('--sample-size', type=int, default=None, help='Sample size (for testing)')
args = parser.parse_args()
output_dir = Path(args.output)
# Load data
print("Loading dataset...")
data_loader = ModelDataLoader()
df = data_loader.load_data(sample_size=args.sample_size)
df = data_loader.preprocess_for_embedding(df)
# Generate embeddings and reduce dimensions if needed
if 'x' not in df.columns or 'y' not in df.columns:
print("Generating embeddings...")
embedder = ModelEmbedder()
embeddings = embedder.generate_embeddings(df['combined_text'].tolist())
print("Reducing dimensions...")
reducer = DimensionReducer()
reduced_embeddings = reducer.reduce_dimensions(embeddings, n_components=3, method='umap')
else:
reduced_embeddings = None
# Export
export_binary_dataset(df, reduced_embeddings, output_dir)
print("Done!")
|