When it comes to League of Legends draft analysis tools, not all are created equal. In this detailed comparison, we'll examine how LoLDraftAI stacks up against DraftGap, another popular draft analysis tool. Through statistical validation, we've found that LoLDraftAI consistently outperforms DraftGap in prediction accuracy (65.6% vs 56.5% on unseen data), thanks to its more sophisticated understanding of league dynamics. This article breaks down the key differences that make LoLDraftAI the superior draft analysis tool.
DraftGap shortcomings
In their own FAQ section, DraftGap explicitly acknowledges their limitations:
Does DraftGap have any shortcomings? DraftGap is not perfect, and there are several things to keep in mind. The overall team comp identity is not taken into account. The synergy of duos within a team are used in the calculations, but the tool does not know about team comp identity like 'engage' or 'poke'. Damage composition is also not used in the calculation (but it is shown, above the team winrate), so you need to keep this in mind on your own. These shortcomings result from the fact that there is not enough data to make a perfect prediction. And we do not want to incorporate opinions like 'malphite is an engage champion' into the tool, as using just data is the most objective way to make a decision.
As we'll demonstrate in this article, LoLDraftAI has overcome these limitations through its advanced machine learning approach, which can identify complex team compositions and their interactions. We will see some examples of when the statistical approach of DraftGap falls short.
DraftGap shortcoming example: full AP draft
Because DraftGap only uses champion pair statistics, it is totally unaware of the draft as a whole. For this reason, it will not understand when a draft only has AP Damage. This can be showcased by creating a full AP Draft with one team having the following champions from top to bot:
- Top: Vladimir
- Jungle: Fiddlesticks
- Middle: Kennen
- Bottom: Heimerdinger
- Support: Taric
When inputting this draft into DraftGap, it predicts a 62.62% win chance. LoLDraftAI on the other hand, understands that this is a full AP Draft, and predicts a win chance of 40.4%.
DraftGap prediction:
LoLDraftAI prediction:
Importantly, this not only impacts analysis but also champion suggestions. For example, against this full AP Draft, LoLDraftAI suggests Ornn as the best toplane champion. DraftGap, on the other hand, thinks that the team with Ornn top against a full AP Draft only has 40% win chance. Obviously Ornn would just be unkillable against a full AP Draft, this is a glaring example of how DraftGap's statistical approach is limited.
Shortcomings conclusion
This full AP draft just serves as a simple illustration, but it also will impact more nuanced situations. DraftGap will not understand:
- Blue vs Red side differences
- When a team has only one or multiple carries
- When a team has no CC
- When a team has low total damage
- When a team only has late game champions
When you add up all these small subtleties, this just makes DraftGap not a very accurate tool, and this is what we will see in the next section that compares the accuracy of DraftGap to LoLDraftAI.
Statistical accuracy comparison
I have assembled a dataset of 5000 games from patch 15.4 to compare the accuracy of DraftGap and LoLDraftAI.
Dataset
The dataset consists of 5000 games from patch 15.4. The games are from EUW ranked solo queue. These are only randomly sampled games that the LoLDraftAI model has not seen during training. The dataset and full results can be found in this google sheet.
Results
Note: When calculating the accuracy, a correct guess is when the side that actually won was predicted to have more than 50% chance of winning.
Overall accuracy:
- DraftGap: 56.46% (2823/5000 correct)
- LoLDraftAI: 65.56% (3278/5000 correct)
Disagreement between models:
- Model Agreement: 3233/5000 samples (64.66%)
- Model Disagreement: 1767/5000 samples (35.34%)
- Accuracy when models agree: 67.03%
- When models disagree, DraftGap accuracy: 37.13%
- When models disagree, LoLDraftAI accuracy: 62.87%
These results demonstrate that LoLDraftAI is significantly more accurate than DraftGap, correctly predicting 65.56% of the games. While it is still impressive that with only a statistical approach, DraftGap can predict 56.46% of the games, it is clear that LoLDraftAI is the superior tool.
Try LoLDraftAI Today
Experience the advanced draft analysis capabilities of LoLDraftAI yourself:
- Use the web version for analysis
- Download the desktop app for live draft tracking
Appendix A: Dataset Verification
The results for DraftGap were obtained by using their source code available here: https://github.com/vigovlugt/draftgap. All results can be manually verified by using their websites and the match id. Example verification for the first match of the dataset: Match results: https://www.leagueofgraphs.com/match/EUW/7298239127
DraftGap prediction:
LoLDraftAI prediction:
Appendix B: Head to Head Comparison
It can be a fun exercise to make the 2 draft tools compete head to head in a draft. I went through this exercise twice, but didn't include it in the main part of the article because it doesn't showcase the differences in such a clear cut manner as the dataset comparison. Both these drafts were obtained by having one tool select champions for one team and the other tool for the other team. The champion with the highest predicted winrate was always picked no matter the role, in the standard competitive pick order(blue first pick, 2 picks red, 2 pick blue etc.).
First draft result
Blue side by DraftGap
- Top: Teemo
- Jungle: Yorick
- Middle: Kennen
- Bottom: Nilah
- Support: Taric
Red side by LoLDraftAI
- Top: Mundo
- Jungle: Sejuani
- Middle: Cho'Gath
- Bottom: Karthus
- Support: Brand
LoLDraftAI predicts an 80% winrate for red side, while DraftGap predicts a 63% winrate for blue side.
While not as clear cut as the full AP example, I think this is another example of how LoLDraftAI outperforms DraftGap. I think red side has comfortable lanes, especially with 2 tanky solo lanes that will be able to rush Magic Resist and be unkillable in lane. The solo lanes are made even harder for blue side by the presence of Karthus ult and ganks from Sejuani.
In my opinion this is also a showcase of how DraftGap can make picks that make sense as pairs, but don't make sense as a whole. In contrast, LoLDraftAI has crafted an original draft that has a lot of tanks but still enough damage to kill the squishy enemy team and where it doesn't matter if the backline of Brand/Karthus is focused, because they can deal their damage no matter what.
DraftGap prediction:
LoLDraftAI prediction:
Second draft result
Blue side by LoLDraftAI
- Top: Vayne
- Jungle: Nunu
- Middle: Cho'Gath
- Bottom: Sivir
- Support: Pyke
Red side by DraftGap
- Top: Malphite
- Jungle: Yorick
- Middle: Kayle
- Bottom: Nilah
- Support: Nami
LoLDraftAI predicts a 65% winrate for blue side, while DraftGap predicts a 67% winrate for red side.
Here again, the models disagree. But I think this reveals another limitation of DraftGap, it is not aware of early/late game dynamics. In this game, I think it is quite easy for the blue team to snowball hard with a pushing mid lane and early gank pressure from Pyke/Nunu. And if they manage to snowball and feed either Vayne/Sivir, they can probably close the game out quickly. This will be even easier because of the objective secures granted by Nunu/Cho'Gath.
DraftGap prediction:
LoLDraftAI prediction:
Conclusion
The head to head comparison, while not as clear cut as the examples in the main article, still can be interpreted as a showcase of how LoLDraftAI is able to understand more nuanced dynamics, rather than just statistical pairings.