:charmandra: 🔥 🔥 🔥

cfg/gru.cfg

@@ -1,27 +1,38 @@
 [net]
-subdivisions=1
 inputs=256
+
+# Test
 batch = 1
+time_steps=1
+
+# Train
+# batch = 512
+# time_steps=64
+
+subdivisions=1
 momentum=0.9
 decay=0.001
-time_steps=1
-learning_rate=0.5
+learning_rate=0.1
 
+burn_in=100
 policy=poly
 power=4
-max_batches=2000
+max_batches=10000
 
 [gru]
 batch_normalize=1
 output = 1024
+tanh = 1
 
 [gru]
 batch_normalize=1
 output = 1024
+tanh = 1
 
 [gru]
 batch_normalize=1
 output = 1024
+tanh = 1
 
 [connected]
 output=256