initial commit

.gitattributes

@@ -0,0 +1 @@
+**/*.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,14 @@
+# Parquet Row Group Eval
+
+A quick experiment to figure out the optimum row group sizing for storing midi data in parquet and accessing it via hugging face.
+
+```
+==================================================
+SUMMARY TABLE
+==================================================
+                         filename  total_rows  num_row_groups  avg_row_group_size  min_row_group_size  max_row_group_size  file_size_mb
+ midi_files_rowgroup_2048.parquet       11262               6                1877                1022                2048        166.41
+ midi_files_rowgroup_4096.parquet       11262               3                3754                3070                4096        166.69
+ midi_files_rowgroup_8192.parquet       11262               2                5631                3070                8192        166.28
+midi_files_rowgroup_16384.parquet       11262               1               11262               11262               11262        166.12
+```

check_parquet_metadata.py

@@ -0,0 +1,79 @@
+#!/usr/bin/env python3
+"""
+Check Parquet file metadata to display row group information.
+"""
+
+import pyarrow.parquet as pq
+from pathlib import Path
+import pandas as pd
+
+# Directory containing the exported files
+parquet_dir = Path("../parquet_row_group_eval")
+
+print("Parquet File Metadata Analysis")
+print("=" * 50)
+
+# Find all parquet files
+parquet_files = list(parquet_dir.glob("*.parquet"))
+parquet_files.sort()
+
+if not parquet_files:
+    print(f"No parquet files found in {parquet_dir}")
+    exit(1)
+
+results = []
+
+for file_path in parquet_files:
+    print(f"\nFile: {file_path.name}")
+    
+    # Read parquet metadata
+    parquet_file = pq.ParquetFile(file_path)
+    metadata = parquet_file.metadata
+    
+    # Extract information
+    num_row_groups = metadata.num_row_groups
+    total_rows = metadata.num_rows
+    file_size = file_path.stat().st_size
+    
+    print(f"  Total rows: {total_rows:,}")
+    print(f"  Number of row groups: {num_row_groups:,}")
+    print(f"  File size: {file_size:,} bytes ({file_size / 1024 / 1024:.2f} MB)")
+    
+    # Get row group details
+    row_group_sizes = []
+    for i in range(num_row_groups):
+        rg = metadata.row_group(i)
+        row_count = rg.num_rows
+        row_group_sizes.append(row_count)
+    
+    avg_row_group_size = sum(row_group_sizes) / len(row_group_sizes) if row_group_sizes else 0
+    min_rg_size = min(row_group_sizes) if row_group_sizes else 0
+    max_rg_size = max(row_group_sizes) if row_group_sizes else 0
+    
+    print(f"  Row group sizes:")
+    print(f"    Average: {avg_row_group_size:.0f} rows")
+    print(f"    Min: {min_rg_size:,} rows")
+    print(f"    Max: {max_rg_size:,} rows")
+    
+    # Store results for summary table
+    results.append({
+        'filename': file_path.name,
+        'total_rows': total_rows,
+        'num_row_groups': num_row_groups,
+        'avg_row_group_size': int(avg_row_group_size),
+        'min_row_group_size': min_rg_size,
+        'max_row_group_size': max_rg_size,
+        'file_size_mb': file_size / 1024 / 1024
+    })
+
+# Summary table
+print("\n" + "=" * 50)
+print("SUMMARY TABLE")
+print("=" * 50)
+
+df = pd.DataFrame(results)
+df = df.sort_values('avg_row_group_size')
+
+print(df.to_string(index=False, float_format='%.2f'))
+
+print(f"\nAnalyzed {len(parquet_files)} files in {parquet_dir}")

export_midi_files_test.py

@@ -0,0 +1,70 @@
+#!/usr/bin/env python3
+"""
+Export MIDI files with different row group sizes for performance testing.
+Only exports MIDI files whose hash key starts with 'a'.
+"""
+
+import duckdb
+import numpy as np
+from pathlib import Path
+import os
+
+# Database path
+db_path = "data/lakh_local.duckdb"
+output_dir = Path("../parquet_row_group_eval")
+
+# Create output directory
+output_dir.mkdir(exist_ok=True)
+
+# Generate 10 row group sizes in logspace from 1024 to 16384
+row_group_sizes = np.logspace(11, 14, base=2, num=4).astype(int)
+print(f"Testing row group sizes: {row_group_sizes}")
+
+# Connect to database
+conn = duckdb.connect(db_path, read_only=True)
+
+# Query to get MIDI files starting with 'a'
+query = """
+SELECT 
+    hm.midi_hk,
+    hm.midi_md5,
+    smf.file_content,
+    smf.file_size,
+    smf.load_date
+FROM hub_midi_file hm
+JOIN sat_midi_file smf ON hm.midi_hk = smf.midi_hk
+WHERE hm.midi_hk LIKE 'a%'
+ORDER BY hm.midi_hk
+"""
+
+print("Querying MIDI files...")
+result = conn.execute(query).fetchdf()
+print(f"Found {len(result)} MIDI files with hash keys starting with 'a'")
+
+# Export with different row group sizes
+for i, row_group_size in enumerate(row_group_sizes):
+    output_file = output_dir / f"midi_files_rowgroup_{row_group_size}.parquet"
+    
+    print(f"Exporting file {i+1}/10: {output_file} (row_group_size={row_group_size})")
+    
+    # Export to parquet with specified row group size
+    conn.execute(f"""
+    COPY (
+        {query}
+    ) TO '{output_file}' 
+    (FORMAT PARQUET, ROW_GROUP_SIZE {row_group_size})
+    """)
+    
+    # Check file size
+    file_size = os.path.getsize(output_file)
+    print(f"  File size: {file_size:,} bytes")
+
+print(f"\nAll files exported to {output_dir}/")
+
+# Show summary
+print("\nSummary of exported files:")
+for file in sorted(output_dir.glob("*.parquet")):
+    size = os.path.getsize(file)
+    print(f"  {file.name}: {size:,} bytes")
+
+conn.close()

midi_files_rowgroup_16384.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a48d08941393d7fae622de4ea7ff8680d384d3fd1d275103f896573e690a98b
+size 174187062

midi_files_rowgroup_2048.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1916da1277760b10c077252e1a60ca9ef963c87895de8f0d8650519b2d95eedf
+size 174494378

midi_files_rowgroup_4096.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ab8dff27d6b3eb49ac8740e6c478d5739897c139d2f5470c55acab16f056c02
+size 174790095

midi_files_rowgroup_8192.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89927aa74f399beb1f1b9783030616af982bd38d996392f744ae87ad7a0c057b
+size 174357122