Merge pull request #371 from macrocosm-os/staging

staging

Author: Sarkosos

README.md

@@ -87,7 +87,7 @@ An ideal incentive mechanism defines an asymmetric workload between the validato
 
 Protein folding is also a research topic that is of incredibly high value. Research groups all over the world dedicate their time to solving particular niches within this space. Providing a solution to attack this problem at scale is what Bittensor is meant to provide to the global community. 
 
-# Simulation Backend and Reproducability
+# Simulation Backend and Reproducibility
 Moleccular dynamics (MD) simulations require a physics-based engine to run them, and SN25 utilizes the open-source project [OpenMM](https://openmm.org). As their tagline suggests, they are a "high performance, customizable molecular simulation" package. 
 
 One of the key advantages of using OpenMM for MD-simulations is the built-in capabilities for *reproducability*. This is a key component in the reward stack and all miners should be intimately familiar with this. For more information, please read this [document](./documentation/reproducibility.md). 

folding/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "2.1.1"
+__version__ = "2.1.2"
 version_split = __version__.split(".")
 __spec_version__ = (
     (10000 * int(version_split[0]))

folding/rewards/linear_reward.py

@@ -4,15 +4,13 @@
 def divide_decreasing(
     amount_to_distribute: float, number_of_elements: int
 ) -> List[float]:
-    # Calculate the fixed decrease amount d
-    d = 2 * amount_to_distribute / ((number_of_elements - 1) * number_of_elements)
+    # Instead of going from n to 0, we go from n to 1
+    # This gives us weights that look like [n, n-1, n-2, ..., 1]
+    weights = [number_of_elements - i for i in range(number_of_elements)]
 
-    # Calculate the first value a1
-    a1 = (
-        amount_to_distribute + d * (number_of_elements - 1) * number_of_elements / 2
-    ) / number_of_elements
+    # Calculate scaling factor to make weights sum to amount_to_distribute
+    total_weight = sum(weights)
+    scaling_factor = amount_to_distribute / total_weight
 
-    # Calculate the n values
-    values = [a1 - i * d for i in range(number_of_elements)]
-
-    return values
+    # Scale all weights proportionally
+    return [w * scaling_factor for w in weights]