Attacks Reference

AI Summary

Complete reference for attack tracking from both TE_Projectile (ranged) and combat log (melee). Captures ALL auto-attacks in the game with consistent fields: entity indices, names, timing, and damage. Typically 30,000+ events per match combining ranged projectiles and melee hits.

---

Overview

The attacks collector unifies two data sources into a single consistent API:

• Ranged attacks: From TE_Projectile messages (projectile speed, dodgeable)
• Melee attacks: From combat log DAMAGE events with no ability (damage values)

Both types share the same AttackEvent model with all fields populated where data is available.

Key Use Cases:

• Last-hit analysis (who attacked which creep)
• Lane farming patterns
• Hero attack tracking (both melee and ranged heroes)
• Tower/creep/neutral combat analysis

---

AttackEvent Fields

Every AttackEvent contains these fields, populated consistently for both ranged and melee:

Timing

Field	Type	Description
tick	int	Game tick (~30/second)
game_time	float	Game time in seconds from horn
game_time_str	str	Formatted time (e.g., "15:34")
launch_tick	int	Tick when projectile was launched (ranged only)

Entity Identification

Field	Type	Ranged	Melee	Description
source_index	int	✓	✓	Entity ID of attacker
target_index	int	✓	✓*	Entity ID of target
attacker_name	str	✓	✓	Unit name (e.g., npc_dota_hero_troll_warlord)
target_name	str	✓	✓	Target unit name

*Melee target_index populated for heroes, may be 0 for creeps/neutrals

Attack Properties

Field	Type	Ranged	Melee	Description
is_melee	bool	False	True	Attack type identifier
damage	int	0	✓	Damage dealt
target_health	int	0	✓	Target health AFTER attack
projectile_speed	int	✓	0	Projectile move speed
dodgeable	bool	✓	False	Can be disjointed
location_x	float	✓	✓*	Attacker position X
location_y	float	✓	✓*	Attacker position Y
attacker_team	int	0	✓	Attacker team (2=Radiant, 3=Dire)
target_team	int	0	✓	Target team (2=Radiant, 3=Dire)
is_attacker_hero	bool	False	✓	Attacker is a hero
is_target_hero	bool	False	✓	Target is a hero

*Melee location from entity lookup (heroes ~37%, creeps 0)

Raw Handles (Advanced)

Field	Type	Description
source_handle	int	Raw entity handle (ranged only)
target_handle	int	Raw entity handle (ranged only)

---

Basic Usage

from python_manta import Parser

parser = Parser("match.dem")

# Collect all attack events (ranged + melee)
result = parser.parse(attacks={})

print(f"Total attacks: {result.attacks.total_events}")

# Separate by type
melee = [a for a in result.attacks.events if a.is_melee]
ranged = [a for a in result.attacks.events if not a.is_melee]

print(f"Melee: {len(melee)}, Ranged: {len(ranged)}")

Limiting Events

# First 1000 attacks only
result = parser.parse(attacks={"max_events": 1000})

---

Hero Name Format

All hero names use the canonical Dota 2 format with underscores:

# Correct format (matches combat log)
"npc_dota_hero_troll_warlord"
"npc_dota_hero_faceless_void"
"npc_dota_hero_shadow_shaman"
"npc_dota_hero_monkey_king"
"npc_dota_hero_storm_spirit"

# NOT: "npc_dota_hero_trollwarlord" (incorrect)

---

Use Case Examples

Lane Farming Analysis

result = parser.parse(attacks={})

# Find all attacks on lane creeps
lane_creep_attacks = [
    a for a in result.attacks.events
    if "creep_goodguys" in a.target_name or "creep_badguys" in a.target_name
]

# Group by attacker
from collections import Counter
attackers = Counter(a.attacker_name for a in lane_creep_attacks)
for name, count in attackers.most_common(10):
    print(f"{name}: {count} attacks on lane creeps")

Contested Last Hits

result = parser.parse(attacks={}, entity_deaths={"include_creeps": True})

# Find creeps that were attacked by multiple heroes before death
from collections import defaultdict

creep_attackers = defaultdict(set)
for attack in result.attacks.events:
    if "npc_dota_hero" in attack.attacker_name:
        creep_attackers[attack.target_name].add(attack.attacker_name)

contested = {k: v for k, v in creep_attackers.items() if len(v) > 1}
print(f"Creeps contested by multiple heroes: {len(contested)}")

Hero vs Hero Combat

result = parser.parse(attacks={})

# Find hero-on-hero attacks
hero_fights = [
    a for a in result.attacks.events
    if "npc_dota_hero" in a.attacker_name and "npc_dota_hero" in a.target_name
]

print(f"Hero vs hero attacks: {len(hero_fights)}")

# Sample output
for a in hero_fights[:5]:
    atk = a.attacker_name.replace("npc_dota_hero_", "")
    tgt = a.target_name.replace("npc_dota_hero_", "")
    dmg = f", dmg={a.damage}" if a.is_melee else ""
    print(f"[{a.game_time_str}] {atk} -> {tgt}{dmg}")

Melee Hero Tracking (e.g., Troll Warlord)

result = parser.parse(attacks={})

# Troll Warlord has both ranged and melee forms
troll_attacks = [a for a in result.attacks.events if "troll_warlord" in a.attacker_name]

troll_melee = [a for a in troll_attacks if a.is_melee]
troll_ranged = [a for a in troll_attacks if not a.is_melee]

print(f"Troll melee attacks: {len(troll_melee)}")
print(f"Troll ranged attacks: {len(troll_ranged)}")

# Melee attacks have damage
if troll_melee:
    avg_dmg = sum(a.damage for a in troll_melee) / len(troll_melee)
    print(f"Average melee damage: {avg_dmg:.1f}")

---

Entity Index Consistency

Entity indices match between attacks and snapshots:

result = parser.parse(attacks={})
snap = parser.snapshot(target_tick=30000)

# Build hero index map
hero_by_id = {h.entity_id: h.hero_name for h in snap.heroes}

# Verify consistency
for attack in result.attacks.events[:100]:
    if attack.source_index in hero_by_id:
        assert attack.attacker_name == hero_by_id[attack.source_index]

---

Statistics (Real Match Data)

From match 8447659831 (Team Spirit vs Tundra, ~46 min):

Metric	Value
Total attacks	32,895
Melee attacks	17,000 (52%)
Ranged attacks	15,895 (48%)
Troll Warlord melee	2,791
Pugna ranged	235

---

Combat Log vs Attacks

Data Type	Combat Log	Attacks Collector
Hero damage	Yes (all)	Yes (auto-attacks only)
Melee auto-attacks	Yes (as damage)	Yes (with entity ID)
Ranged auto-attacks	No	Yes (with projectile data)
Tower attacks on creeps	No	Yes
Neutral attacks	No	Yes
Ability damage	Yes	No
Kill attribution	Yes	No

Rule of thumb: - Use combat log for damage analysis and kill tracking - Use attacks for attack pattern analysis and entity correlation

---

Performance Notes

• Total events: ~30,000+ per match (ranged + melee)
• Use max_events to limit for faster parsing
• Events are ordered by tick
• Entity name → index mapping is automatic for heroes