High Value Project Tutorials - Cost Reduction Information
Cost Reduction Factors
| Cost Reduction Factor | Without DataPredict | With DataPredict |
|---|---|---|
| Expertise | Requires Hiring Of A Machine Learning Engineer, Starting At 5K USD Per Month To Build Optimized Models Required By Your Projects. Preferrably Holding A PhD Or Masters Degree To Ensure Correctness. | Can Use Existing Game Developers. Can Learn From DataPredict’s High-Value Project Tutorials To Specialize Further. |
| Integration | Additional Costs From Hosting External Servers And Connecting ML Frameworks In Different Programming Languages. | Lua Only. |
| Compute Cost | If Models Are Hosted Externally, Compute Cost Increases As Number Of Players Increases. | Utilize Roblox Servers’ And Players’ CPUs At No Additional Cost. |
| Choosing Models | Need To Learn Multiple ML Frameworks For Machine Learning, Deep Learning And Reinforcement Learning In Different Programming Languages If Models Are Hosted Externally. | Unified Under Same API Design. All Models With High-Value Use Cases Are Covered. |
| Time-To-Market | Requires Multi-Year Engineering For The Whole Library If Building From Scratch. Minimum Of 1 Week If Using Existing ML Frameworks Hosted In Different Programming Languages. | Within A Day. |
| Latency | Network Latency And Serialization Issues If Hosted Externally. | Latency Only Comes From Models’ Training And Predictions. Mostly Depends On Roblox’s Servers And Clients. |
| Security | If Hosted Externally, Vulnerable To Hacking. | Data Stays Between The Roblox’s Servers And Clients. |
| Reliability | May Be Inaccurate Or Underperform When Building From Scratch. If Hosted Externally, Models May Be Impacted By Third Party Servers. | 3-Years Of Development. |
| Legal | If Hosted Externally, Requires Active Compliance For Data Protection Laws. | Data Stays Between The Servers And Clients. As A Result, Mostly Complies With Data Protection laws. |
| Domain Specialization | Machine Learning Engineers Generally Do Not Specialize In Games. | Built With Game Domain Specialties And Optimizations. |
Licensing Costs
| License Type Based On Gross Revenue Per Company / Organization / Group / Individual | Percentage To Pay Based On Project’s Gross Revenue In USD | Additional Notes | Licensing Cost When A Project Earns 5K / 10K / 100K / 1M USD Per Month |
|---|---|---|---|
| Non-B2B + Less Than Or Equal To 5K USD Within 365 Days (Not Per 365 Days) | 0% | Requires Public Disclosure Of DataPredict | Not Applicable |
| Standard | 2% That Decreases As Gross Revenue Increases | Requires Public Disclosure Of DataPredict | 89 USD / 160 USD / 1300 USD / 10,530 USD |
| White-Label | 5% | None | 250 USD / 500 USD / 5000 USD / 50,000 USD |
- Note: At the $5,000/year revenue mark, the licensing fee is approximately $8/month for “Standard” license. Collecting micro-payments of this size creates administrative overhead that exceeds the revenue itself. Therefore, we subsidize 100% of costs for organizations under this threshold. We only begin billing when your success generates a fee large enough to be meaningful for both parties.
Development Costs
The table assumes that only one person is working on each library at machine learning engineer average salary rate in the United States.
| Library | Personal Time Cost | Work Hours Time Cost (40 Hours Per Week) | Personal Monetary Cost (USD) | Work Hours Monetary Cost (USD) |
|---|---|---|---|---|
| DataPredict™ | 25000+ Hours (Spread Over 3 Years) | 625 Weeks (~12 Years) | ~300,000 | ~1,260,000 |
| DataPredict™ Neural | 17000+ Hours (Spread Over 2 Years) | 425 Weeks (~8.2 Years) | ~200,000 | ~840,000 |
| DataPredict™ Axon | 4000+ Hours (Spread Over 6 Months) | 100 Weeks (~2 Years) | ~50,000 | ~210,000 |
| DataPredict™ Zero | 2000+ Hours (Spread Over 3 Months) | 50 Weeks (~1 Year) | ~25,000 | ~105,000 |
| All | 48000+ Hours (Spread Over 5 Years) | 1095 Weeks (~21 years) | ~575,000 | ~2,204,000 |
This assumes that the machine learning engineer work at these timing:
-
Weekdays: 7 pm to 3 am. Roughly 8 hours of work.
-
Weekends: 9 am to 3 am. Roughly 18 hours of work.
-
Weekly Total: Roughly 76 hours of work.
Factors That Led To These Pricing Strategies
-
DataPredict was initially built from passion as a fresh graduate, but eventually went to commercial licensing to allow it to be capable for production-grade use. Hence, the first year of development costs are considered “subsidized” due to it being more of a learning experience more than building a serious project.
-
I am based from Malaysia, and I can keep the costs low compared to the United States.
-
Being a solo developer means that there is no overhead from managing people, keeping the costs low.
-
I rely on the community to search for potential bugs in addition of my own bug hunting, keeping the costs low by not hiring a dedicated tester.
To understand why I don’t use AIs to let the AI to do full code generation to further lower the costs, you can have a look here.
Comparison To The Solution Of Other Game Engines
| ML For Games Solution | Description | Features | First Development Date | First Release Date | Development Duration For Production Release |
|---|---|---|---|---|---|
| Unity’s ML-Agents | Uses Unity SDK, PyTorch And External Communicator To Train AI Agents. | PPO, SAC And Imitation Learning. | 19 September 2017 | 1 May 2020 | 955 days (2.7 Years) |
| Unreal Engine’s Learning Agents | Uses Unreal Engine’s Blueprint System And C++ library To Train AI Agents. | PPO And Behavior Cloning. | September 2023 | Still At Experimental Phase | N/A |
| Godot’s RL Agents (Unofficial) | A Wrapper For StableBaselines3, Sample Factory, Ray RLLib And CleanRL To Train AI Agents. | A Variety Of Reinforcement Learning Models. | 18 October 2021 | Stalled At 26 February 2025 | N/A |