major cleanup in rate distortion

README.md

@@ -27,6 +27,10 @@ First, set up a virtual environment (e.g. via [miniconda](https://docs.conda.io/
 
     `pip install -e .`
 
+3. In addition, this project requires [rdot](), a python library of rate-distortion optimization tools. When a stable version is available, we will add this to the ULTK `setup.py` file; for now, install via git:
+
+    `python3 -m pip install git+https://github.com/nathimel/rdot.git`
+
 ## Getting started
 
 - Check out the [examples](https://github.com/CLMBRs/ultk/tree/main/src/examples), starting with a basic signaling game.  The examples folder also contains a simiple efficient communication analysis of [indefinites](https://github.com/CLMBRs/ultk/tree/main/src/examples/indefinites).

src/ultk/effcomm/agent.py

@@ -296,24 +296,3 @@ def __init__(
         for i in range(len(self.R)):
             col = speaker.S[:, i]
             self.R[i] = col @ prior / np.sum(col @ prior)
-
-
-class BayesianListener(Listener):
-    """A Bayesian reciever chooses an interpretation according to p(meaning | word), where
-
-    # BUG: This is extremely misleading since we basically only use this function for IB, and IB assumes a DETERMINISTIC bayes-derived listener.
-
-    $P(m | w) = \\frac{P(M | W) \cdot P(M)} { P(W) }$
-
-    Furthermore, we sometimes require that each word w is deterministically interpreted as meaning $\hat{m}$ as follows:
-
-    # BUG: This says nothing about determinism.
-    $\hat{m}_{w}(u) = \sum_m p(m|w) \cdot m(u)$
-
-    See ultk.effcomm.information for more details.
-    """
-
-    def __init__(self, speaker: Speaker, prior: np.ndarray, name: str = None):
-        weights = bayes(speaker.normalized_weights(), prior)
-        # TODO: Change this whole class to DeterministicBayesOptimalListener, and implement the correct weights!
-        super().__init__(speaker.language, weights=weights, name=name)