comparison sargo

SainsburyWellcomeCentre · Jul 15, 2023 · b6c6c52 · b6c6c52
1 parent e4eaaf9
commit b6c6c52
Show file tree

Hide file tree

Showing 3 changed files with 164 additions and 10,122 deletions.
diff --git a/examples/comparisons_examples/Comparision_from_manadger.ipynb b/examples/comparisons_examples/Comparision_from_manadger.ipynb
diff --git a/neuralplayground/agents/stachenfeld_2018.py b/neuralplayground/agents/stachenfeld_2018.py
@@ -396,41 +396,6 @@ def plot_transition(self, save_path: str = None, ax: mpl.axes.Axes = None):
             plt.close("all")
         return ax
 
-    def plot_eigen(
-        self,
-        matrix: np.ndarray,
-        save_path: str,
-        eigen_vectors=(0, 1),
-        ax: mpl.axes.Axes = None,
-    ):
-        """ "
-        Plot the matrix and the 4 largest modes of its eigen-decomposition
-
-        Parameters
-        ----------
-        matrix: array
-            The matrix that will be plotted
-        eigen:  np.ndarray
-            Which eigenvectors you would like to plot
-        save_path: string
-            Path to save the plot
-        """
-        evals, evecs = np.linalg.eig(matrix)
-        if ax is None:
-            f, ax = plt.subplots(1, len(eigen_vectors), figsize=(4 * len(eigen_vectors), 5))
-            if len(eigen_vectors) == 1:
-                evecs_0 = evecs[:, eigen_vectors[0]].reshape(self.depth, self.width).real
-                make_plot_rate_map(evecs_0, ax, "Eig_0", "width", "depth", "Firing rate")
-            else:
-                for i, eig in enumerate(eigen_vectors):
-                    evecs_0 = evecs[:, eig].reshape(self.depth, self.width).real
-                    make_plot_rate_map(evecs_0, ax[i], "Eig" + str(eig), "width", "depth", "Firing rate")
-        if save_path is None:
-            pass
-        else:
-            plt.savefig(save_path, bbox_inches="tight")
-        return ax
-
     def get_rate_map_matrix(
         self,
         sr_matrix,

diff --git a/neuralplayground/backend/training_loops.py b/neuralplayground/backend/training_loops.py
@@ -26,13 +26,15 @@ def default_training_loop(agent: AgentCore, env: Environment, n_steps: int):
 
     obs, state = env.reset()
     training_hist = []
+    obs = obs[:2]
     for j in range(round(n_steps)):
         # Observe to choose an action
         action = agent.act(obs)
         # Run environment for given action
         obs, state, reward = env.step(action)
         update_output = agent.update()
         training_hist.append(update_output)
+        obs = obs[:2]
     process_training_hist(training_hist)
     return agent, env, dict