From a01d83f33996d2d976492772465ee03570f780c6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Matt=C3=A9o=20Delabre?= <spam@delab.re>
Date: Fri, 13 Dec 2019 17:10:40 -0500
Subject: [PATCH] Improve plot

---
 plot.py  | 77 +++++++++++++++++++++++++++++++++++++++++++++++++++++---
 train.py | 27 ++++++++++----------
 2 files changed, 87 insertions(+), 17 deletions(-)

diff --git a/plot.py b/plot.py
index 7c0bd9b..65d9041 100644
--- a/plot.py
+++ b/plot.py
@@ -2,12 +2,15 @@ import constellation
 from constellation import util
 import torch
 from matplotlib import pyplot
+from mpl_toolkits.axisartist.axislines import SubplotZero
+import math
+import numpy
 
 # Number learned symbols
 order = 4
 
 # File in which the trained model is saved
-input_file = 'output/constellation-net.tc'
+input_file = 'output/constellation-order-{}.tc'.format(order)
 
 model = constellation.ConstellationNet(order=order)
 model.load_state_dict(torch.load(input_file))
@@ -19,8 +22,74 @@ with torch.no_grad():
             torch.arange(0, order),
             order
         )
-    ).tolist()
+    )
+
+fig = pyplot.figure()
+ax = SubplotZero(fig, 111)
+fig.add_subplot(ax)
+
+# Extend axes symmetrically around zero so that they fit data
+extent = max(
+    abs(encoded_vectors.min()),
+    abs(encoded_vectors.max())
+) * 1.05
+
+ax.set_xlim(-extent, extent)
+ax.set_ylim(-extent, extent)
+
+# Hide borders
+for direction in ['left', 'bottom', 'right', 'top']:
+    ax.axis[direction].set_visible(False)
+
+# Show zero-centered axes
+for direction in ['xzero', 'yzero']:
+    axis = ax.axis[direction]
+    axis.set_visible(True)
+    axis.set_axisline_style("-|>")
+
+# Configure axes ticks and labels
+ax.annotate(
+    'I', (1, 0.5), xycoords='axes fraction',
+    xytext=(25, 0), textcoords='offset points',
+    va='center', ha='right'
+)
+
+ax.axis['xzero'].major_ticklabels.set_backgroundcolor('white')
+ax.axis['xzero'].major_ticklabels.set_ha('center')
+ax.axis['xzero'].major_ticklabels.set_va('top')
+
+ax.annotate(
+    'Q', (0.5, 1), xycoords='axes fraction',
+    xytext=(0, 25), textcoords='offset points',
+    va='center', ha='center'
+)
+
+ax.axis['yzero'].major_ticklabels.set_rotation(-90)
+ax.axis['yzero'].major_ticklabels.set_backgroundcolor('white')
+ax.axis['yzero'].major_ticklabels.set_ha('left')
+ax.axis['yzero'].major_ticklabels.set_va('center')
+
+# Add a single tick on 0
+ax.set_xticks(ax.get_xticks()[ax.get_xticks() != 0])
+ax.set_yticks(ax.get_yticks()[ax.get_yticks() != 0])
+
+ax.annotate(
+    '0', (0, 0),
+    xytext=(15, -10), textcoords='offset points',
+    va='center', ha='center'
+)
+
+ax.grid()
+
+# Plot encoded vectors
+ax.scatter(*zip(*encoded_vectors.tolist()), zorder=10)
+
+# Add index label for each vector
+for row in range(order):
+    ax.annotate(
+        row + 1, encoded_vectors[row],
+        xytext=(5, 5), textcoords='offset points',
+        backgroundcolor='white', zorder=9
+    )
 
-fig, axis = pyplot.subplots()
-axis.scatter(*zip(*encoded_vectors))
 pyplot.show()
diff --git a/train.py b/train.py
index 214bc11..3d1eacd 100644
--- a/train.py
+++ b/train.py
@@ -15,11 +15,14 @@ num_epochs = 20000
 loss_report_epoch_skip = 500
 
 # File in which the trained model is saved
-output_file = 'output/constellation-net.tc'
-
-print('Starting training with {} epochs\n'.format(num_epochs))
+output_file = 'output/constellation-order-{}.tc'.format(order)
 
 model = constellation.ConstellationNet(order=order)
+
+# Train the model with random data
+model.train()
+print('Starting training with {} epochs\n'.format(num_epochs))
+
 criterion = torch.nn.CrossEntropyLoss()
 optimizer = torch.optim.Adam(model.parameters())
 
@@ -46,24 +49,22 @@ for epoch in range(num_epochs):
 print('\nFinished training\n')
 
 # Print some examples of reconstruction
-with torch.no_grad():
-    num_examples = 5
+model.eval()
+print('Reconstruction examples:')
+print('Input vector\t\t\tOutput vector after softmax')
 
-    classes_example = util.get_random_messages(num_examples, order)
-    onehot_example = util.messages_to_onehot(classes_example, order)
+with torch.no_grad():
+    onehot_example = util.messages_to_onehot(torch.arange(0, order), order)
     raw_output = model(onehot_example)
     raw_output.required_grad = False
     reconstructed_example = torch.nn.functional.softmax(raw_output, dim=1)
 
-    print('Reconstruction examples:')
-    print('Input vector\t\t\tOutput vector after softmax')
-
-    for example_index in range(num_examples):
+    for index in range(order):
         print('{}\t\t{}'.format(
-            onehot_example[example_index].tolist(),
+            onehot_example[index].tolist(),
             '[{}]'.format(', '.join(
                 '{:.5f}'.format(x)
-                for x in reconstructed_example[example_index].tolist()
+                for x in reconstructed_example[index].tolist()
             ))
         ))