github-starred/TinyTorch
2
0
Fork 0
mirror of https://github.com/MLSysBook/TinyTorch.git synced 2025-12-05 19:17:52 -06:00
Code Issues 7 Packages Projects Releases Wiki Activity

Add table support to ASCII box fixer and fix table alignment

Browse Source 
- Add table detection (┬ ┼ ┴ column separators)
- Fix table alignment by adjusting cell widths
- Flag tables with content wider than headers for manual review
- Manually fix tables in 04_losses.py (expanded column widths)
- Fix table in 01_tensor.py
This commit is contained in:
Vijay Janapa Reddi
2025-11-30 09:48:30 -05:00
parent c4d0bdb901
commit 7bd0210324
3 changed files with 277 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/01_tensor/01_tensor.py
View File

@@ -834,7 +834,7 @@ Neural Network Usage:
┌─────────────────────┬─────────────────────┬─────────────────────┐
│ Weight Matrices │ Attention Mechanism │ Gradient Computation│
├─────────────────────┼─────────────────────┼─────────────────────┤
│ Forward: X @ W │ Q @ K^T attention │ ∂L/∂W = X^T @ ∂L/∂Y│
│ Forward: X @ W │ Q @ K^T attention │ ∂L/∂W = X^T @ ∂L/∂Y │
│ Backward: X @ W^T │ scores │ │
└─────────────────────┴─────────────────────┴─────────────────────┘
```

28
src/04_losses/04_losses.py
View File

@@ -532,14 +532,14 @@ Loss: -log(0.003) = 5.8 ← Very high loss ❌
```
What Cross-Entropy Teaches the Model:
┌─────────────────┬─────────────────┬─────────────────┐
│ Prediction │ True Label │ Learning Signal │
├─────────────────┼─────────────────┼─────────────────┤
│ Confident ✅ │ Correct ✅ │ "Keep doing this"│
│ Uncertain ⚠️ │ Correct ✅ │ "Be more confident"│
│ Confident ❌ │ Wrong ❌ │ "STOP! Change everything"│
│ Uncertain ⚠️ │ Wrong ❌ │ "Learn the right answer"│
└─────────────────┴─────────────────┴─────────────────┘
┌─────────────────┬─────────────────┬───────────────────────────┐
│ Prediction │ True Label │ Learning Signal │
├─────────────────┼─────────────────┼───────────────────────────┤
│ Confident ✅ │ Correct ✅ │ "Keep doing this" │
│ Uncertain ⚠️ │ Correct ✅ │ "Be more confident" │
│ Confident ❌ │ Wrong ❌ │ "STOP! Change everything" │
│ Uncertain ⚠️ │ Wrong ❌ │ "Learn the right answer" │
└─────────────────┴─────────────────┴───────────────────────────┘
Loss Landscape by Confidence:
Loss
@@ -1043,26 +1043,26 @@ Different loss functions have different computational costs, especially at scale
Computational Cost Comparison (Batch Size B, Classes C):
MSELoss:
┌───────────────┬───────────────┐
┌────────────────┬────────────────┐
│ Operation │ Complexity │
├───────────────┼───────────────┤
├────────────────┼────────────────┤
│ Subtraction │ O(B) │
│ Squaring │ O(B) │
│ Mean │ O(B) │
│ Total │ O(B) │
└───────────────┴───────────────┘
└────────────────┴────────────────┘
CrossEntropyLoss:
┌───────────────┬───────────────┐
┌────────────────┬────────────────┐
│ Operation │ Complexity │
├───────────────┼───────────────┤
├────────────────┼────────────────┤
│ Max (stability)│ O(B*C) │
│ Exponential │ O(B*C) │
│ Sum │ O(B*C) │
│ Log │ O(B) │
│ Indexing │ O(B) │
│ Total │ O(B*C) │
└───────────────┴───────────────┘
└────────────────┴────────────────┘
Cross-entropy is C times more expensive than MSE!
For ImageNet (C=1000), CE is 1000x more expensive than MSE.

286
tools/dev/fix_ascii_boxes.py
View File

@@ -1,23 +1,24 @@
#!/usr/bin/env python3
"""
Fix ASCII Box Alignment
Fix ASCII Box and Table Alignment
This script finds simple ASCII art boxes in Python files and ensures the
This script finds ASCII art boxes and tables in Python files and ensures the
right-side vertical bars (│) are perfectly aligned with the top border.
Handles:
- Simple boxes (content lines with exactly 2 │)
- Boxes with ├───┤ separator lines
- Tables with columns (┬ ┼ ┴ separators)
Skips (requires manual review):
- Nested boxes (content lines with more than 2 │)
- Nested boxes (boxes inside boxes)
- Side-by-side boxes
- Dashed boxes
Usage:
python tools/fix_ascii_boxes.py # Preview changes (dry run)
python tools/fix_ascii_boxes.py --fix # Apply fixes
python tools/fix_ascii_boxes.py --verbose # Show detailed info
python tools/dev/fix_ascii_boxes.py # Preview changes (dry run)
python tools/dev/fix_ascii_boxes.py --fix # Apply fixes
python tools/dev/fix_ascii_boxes.py --verbose # Show detailed info
"""
import sys
@@ -185,6 +186,185 @@ def fix_box_alignment(box_lines: list[str]) -> list[str]:
return fixed_lines
def find_tables(content: str) -> list[tuple[int, int, list[str]]]:
"""
Find ASCII tables (boxes with column separators ┬ ┼ ┴).
Returns list of (start_line, end_line, lines) tuples.
"""
lines = content.split('\n')
tables = []
i = 0
while i < len(lines):
line = lines[i]
# Look for table start: ┌...┬...┐ (has column separator)
if '┌' in line and '┐' in line and '┬' in line:
first_corner = line.index('┌')
# Skip nested (│ before ┌)
if '│' in line[:first_corner]:
i += 1
continue
table_start = i
table_lines = [line]
left_pos = first_corner
# Collect table lines
j = i + 1
while j < len(lines) and j - i < 100:
current_line = lines[j]
# Content line (multiple │)
if '│' in current_line:
first_bar = current_line.index('│')
if first_bar == left_pos:
table_lines.append(current_line)
j += 1
else:
break
# Separator line with ┼
elif '├' in current_line and '┤' in current_line:
table_lines.append(current_line)
j += 1
# Bottom line with ┴
elif '└' in current_line and '┘' in current_line:
table_lines.append(current_line)
tables.append((table_start, j, table_lines))
i = j
break
else:
break
i += 1
else:
i += 1
return tables
def fix_table_alignment(table_lines: list[str]) -> list[str]:
"""
Fix alignment of a table with columns.
Strategy: Find the column separator positions from the TOP line,
then adjust each content row's cells to match those widths.
"""
if len(table_lines) < 3:
return table_lines
top_line = table_lines[0]
left_pos = top_line.index('┌')
right_pos = top_line.index('┐')
total_width = right_pos - left_pos + 1
# Find column separator positions (┬) in top line
separators = [i for i, c in enumerate(top_line) if c == '┬']
# Column boundaries: [left_pos, sep1, sep2, ..., right_pos]
boundaries = [left_pos] + separators + [right_pos]
# Calculate column widths (between separators)
col_widths = [boundaries[i+1] - boundaries[i] - 1 for i in range(len(boundaries) - 1)]
fixed_lines = [top_line] # Top line stays as-is (defines structure)
for line in table_lines[1:-1]:
prefix = line[:left_pos]
if '├' in line and '┤' in line:
# Separator row - rebuild with ┼
parts = ['├']
for i, w in enumerate(col_widths):
parts.append('─' * w)
if i < len(col_widths) - 1:
parts.append('┼')
parts.append('┤')
fixed_lines.append(prefix + ''.join(parts))
elif '│' in line:
# Content row - extract cells by splitting on │
# Remove prefix, get content between first and last │
first_bar = line.index('│')
last_bar = line.rindex('│')
inner = line[first_bar + 1:last_bar]
# Split on │ to get cells
cells = inner.split('│')
# Pad/trim each cell to its column width
fixed_cells = []
for i, cell in enumerate(cells):
if i < len(col_widths):
w = col_widths[i]
# Preserve leading space, trim/pad to width
cell_stripped = cell.rstrip()
if len(cell_stripped) > w:
cell_stripped = cell_stripped[:w]
fixed_cells.append(cell_stripped.ljust(w))
else:
fixed_cells.append(cell)
fixed_line = prefix + '│' + '│'.join(fixed_cells) + '│'
fixed_lines.append(fixed_line)
else:
fixed_lines.append(line)
# Bottom line - rebuild with ┴
bottom_line = table_lines[-1]
if '└' in bottom_line and '┘' in bottom_line:
prefix = bottom_line[:left_pos]
parts = ['└']
for i, w in enumerate(col_widths):
parts.append('─' * w)
if i < len(col_widths) - 1:
parts.append('┴')
parts.append('┘')
fixed_lines.append(prefix + ''.join(parts))
else:
fixed_lines.append(bottom_line)
return fixed_lines
def table_needs_fixing(table_lines: list[str]) -> tuple[bool, bool]:
"""
Check if table has misaligned columns.
Returns (needs_fixing, content_too_wide).
content_too_wide means content is wider than header - needs manual fix.
"""
if len(table_lines) < 3:
return False, False
top_line = table_lines[0]
left_pos = top_line.index('┌')
right_pos = top_line.index('┐')
content_too_wide = False
misaligned = False
for line in table_lines[1:]:
if '│' in line:
last_bar = line.rindex('│')
if last_bar != right_pos:
misaligned = True
if last_bar > right_pos:
content_too_wide = True
elif '┤' in line:
last_corner = line.rindex('┤')
if last_corner != right_pos:
misaligned = True
elif '┘' in line:
last_corner = line.rindex('┘')
if last_corner != right_pos:
misaligned = True
return misaligned, content_too_wide
def count_complex_boxes(content: str) -> int:
"""Count boxes that are too complex to auto-fix."""
lines = content.split('\n')
@@ -195,6 +375,10 @@ def count_complex_boxes(content: str) -> int:
line = lines[i]
if '┌' in line and '┐' in line:
first_corner = line.index('┌')
# Skip tables (handled separately)
if '┬' in line:
i += 1
continue
# Nested box
if '│' in line[:first_corner]:
count += 1
@@ -208,23 +392,26 @@ def count_complex_boxes(content: str) -> int:
return count
def process_file(filepath: Path, fix: bool = False, verbose: bool = False) -> tuple[bool, int, int]:
"""Process a single file. Returns (has_changes, num_fixed, num_complex)."""
def process_file(filepath: Path, fix: bool = False, verbose: bool = False) -> tuple[bool, int, int, int]:
"""Process a single file. Returns (has_changes, num_boxes_fixed, num_tables_fixed, num_complex)."""
try:
content = filepath.read_text(encoding='utf-8')
except Exception as e:
return False, 0, 0
return False, 0, 0, 0
boxes = find_simple_boxes(content)
tables = find_tables(content)
complex_count = count_complex_boxes(content)
if not boxes and complex_count == 0:
return False, 0, 0
if not boxes and not tables and complex_count == 0:
return False, 0, 0, 0
lines = content.split('\n')
original_content = content
boxes_fixed = 0
tables_fixed = 0
# Fix boxes
for start_line, end_line, box_lines in reversed(boxes):
if not needs_fixing(box_lines):
continue
@@ -249,13 +436,53 @@ def process_file(filepath: Path, fix: bool = False, verbose: bool = False) -> tu
lines[start_line:end_line + 1] = fixed_lines
# Rebuild content after box fixes
content = '\n'.join(lines)
lines = content.split('\n')
# Find tables again (line numbers may have shifted)
tables = find_tables(content)
# Fix tables
tables_skipped = 0
for start_line, end_line, table_lines in reversed(tables):
misaligned, content_too_wide = table_needs_fixing(table_lines)
if not misaligned:
continue
if content_too_wide:
# Content is wider than header - needs manual fix
tables_skipped += 1
if verbose:
print(f"\n ⚠️ Table at lines {start_line + 1}-{end_line + 1} has content wider than header (manual fix needed)")
continue
fixed_lines = fix_table_alignment(table_lines)
if fixed_lines != table_lines:
tables_fixed += 1
if verbose:
print(f"\n 📊 Table at lines {start_line + 1}-{end_line + 1}:")
print(" Before:")
for line in table_lines:
print(f" {line}")
print(" After:")
for line in fixed_lines:
print(f" {line}")
lines[start_line:end_line + 1] = fixed_lines
complex_count += tables_skipped # Count tables needing manual fix as complex
new_content = '\n'.join(lines)
changes_made = new_content != original_content
if changes_made and fix:
filepath.write_text(new_content, encoding='utf-8')
return changes_made, boxes_fixed, complex_count
return changes_made, boxes_fixed, tables_fixed, complex_count
def main():
@@ -286,36 +513,47 @@ def main():
print(f"🔍 Scanning {len(py_files)} Python files...\n")
total_fixed = 0
total_boxes = 0
total_tables = 0
total_complex = 0
files_changed = 0
for filepath in sorted(py_files):
has_changes, fixed, complex_count = process_file(filepath, fix=args.fix, verbose=args.verbose)
total_fixed += fixed
has_changes, boxes_fixed, tables_fixed, complex_count = process_file(filepath, fix=args.fix, verbose=args.verbose)
total_boxes += boxes_fixed
total_tables += tables_fixed
total_complex += complex_count
total_fixed = boxes_fixed + tables_fixed
if has_changes or complex_count > 0:
if has_changes:
files_changed += 1
status = "✅ Fixed" if args.fix and fixed > 0 else "⚠️ Needs fixing" if fixed > 0 else ""
status = "✅ Fixed" if args.fix and total_fixed > 0 else "⚠️ Needs fixing" if total_fixed > 0 else ""
parts = []
if fixed > 0:
parts.append(f"{fixed} simple")
if boxes_fixed > 0:
parts.append(f"{boxes_fixed} box{'es' if boxes_fixed != 1 else ''}")
if tables_fixed > 0:
parts.append(f"{tables_fixed} table{'s' if tables_fixed != 1 else ''}")
if complex_count > 0:
parts.append(f"{complex_count} complex (manual)")
parts.append(f"{complex_count} complex")
if parts:
print(f"{status}: {filepath} ({', '.join(parts)})" if status else f"📋 {filepath} ({', '.join(parts)})")
print()
if total_fixed == 0 and total_complex == 0:
print("✨ All ASCII boxes are properly aligned.")
total_all = total_boxes + total_tables
if total_all == 0 and total_complex == 0:
print("✨ All ASCII boxes and tables are properly aligned.")
else:
if total_fixed > 0:
if total_all > 0:
if args.fix:
print(f"✅ Fixed {total_fixed} simple box{'es' if total_fixed != 1 else ''} in {files_changed} file{'s' if files_changed != 1 else ''}.")
parts = []
if total_boxes > 0:
parts.append(f"{total_boxes} box{'es' if total_boxes != 1 else ''}")
if total_tables > 0:
parts.append(f"{total_tables} table{'s' if total_tables != 1 else ''}")
print(f"✅ Fixed {' and '.join(parts)} in {files_changed} file{'s' if files_changed != 1 else ''}.")
else:
print(f"⚠️ Found {total_fixed} misaligned simple box{'es' if total_fixed != 1 else ''}.")
print(f"⚠️ Found {total_all} misaligned item{'s' if total_all != 1 else ''}.")
print(" Run with --fix to apply corrections.")
if total_complex > 0:
print(f"📋 Found {total_complex} complex/nested box{'es' if total_complex != 1 else ''} (require manual review).")