docs, notes, names

dominoes.ipynb

@@ -134,7 +134,12 @@
     "\n",
     "# TSP Distance Traveled:\n",
     "# - Explicitly measure the distance traveled by the agent in the TSP task\n",
-    "# - Compare to Held-Karp Solution\n"
+    "# - Compare to Held-Karp Solution\n",
+    "\n",
+    "\n",
+    "# TODO ASAP!!!!!!\n",
+    "# - Get test result plots in there for good plotting, then start running experiments with different parameters\n",
+    "#   so you can save / see the results!!!\n"
    ]
   },
   {

dominoes/experiments/ptr_arch_comp.py

@@ -6,7 +6,7 @@
 from ..networks import get_pointer_network, get_pointer_methods, get_pointer_arguments
 from .base import Experiment
 from . import arglib
-from ..plotting import plot_basic_results
+from ..plotting import plot_train_results
 
 
 class PointerArchitectureComparison(Experiment):
@@ -117,4 +117,4 @@ def plot(self, results):
         main plotting loop
         """
         pointer_methods = self.pointer_methods()
-        plot_basic_results(results["train_results"], pointer_methods, train=True)
+        plot_train_results(self, results["train_results"], pointer_methods, name="training")

dominoes/plotting.py

@@ -5,6 +5,15 @@
 
 
 def plot_train_results(exp, results, labels, name="train"):
+    """
+    Simplest plot method for plotting loss/reward (or something else eventually) across epochs
+
+    Assumes that the loss/rewards are divided into len(labels) types across the first dimension
+    (see below for variable names), will make a plot of the mean across epochs for each type and
+    label it accordingly.
+
+    The experiment object passed in as the first argument determines if the plot is saved or shown.
+    """
     num_types = len(labels)
 
     if "loss" in results and results["loss"] is not None: