2019赛季RM中部分区赛，自瞄和能量机关，完整稳定版。

tools/TrainCNN/backward.py

@@ -75,7 +75,7 @@ def train(dataset, show_bar=False):
     learning_rate = tf.train.exponential_decay(
         LEARNING_RATE_BASE,
         global_step,
-        len(dataset.train_samples) / BATCH / 5,
+        len(dataset.train_samples) / BATCH,
         LEARNING_RATE_DECAY,
         staircase=False)
     train_step = tf.train.AdamOptimizer(learning_rate).minimize(loss, global_step=global_step)

tools/TrainCNN/forward.py

@@ -71,16 +71,14 @@ def forward(x, regularizer=None):
     pool_shape = pool2.get_shape().as_list()
     node = pool_shape[1] * pool_shape[2] * pool_shape[3]
     reshaped = tf.reshape(pool2, [-1, node])
-#    reshaped = tf.nn.dropout(reshaped, 0.1)
 
-    fc1_w = get_weight([node, FC1_OUTPUT_NODES], regularizer)
+    fc1_w = tf.nn.dropout(get_weight([node, FC1_OUTPUT_NODES], regularizer), 0.1)
     fc1_b = get_bias([FC1_OUTPUT_NODES])
     fc1   = tf.nn.relu(tf.matmul(reshaped, fc1_w) + fc1_b)
-#    fc1   = tf.nn.dropout(fc1, 0.2)
     vars.extend([fc1_w, fc1_b])
     nodes.extend([fc1])
 
-    fc2_w = get_weight([FC1_OUTPUT_NODES, FC2_OUTPUT_NODES], regularizer)
+    fc2_w = tf.nn.dropout(get_weight([FC1_OUTPUT_NODES, FC2_OUTPUT_NODES], regularizer), 0.1)
     fc2_b = get_bias([FC2_OUTPUT_NODES])
     fc2   = tf.nn.softmax(tf.matmul(fc1, fc2_w) + fc2_b)
     vars.extend([fc2_w, fc2_b])

tools/TrainCNN/generate.py

@@ -41,13 +41,6 @@ class DataSet:
 
     def generate_data_sets(self, folder):
         sets = []
-        mini = 1000000
-        for i in range(OUTPUT_NODES):
-            dir = "%s/%d" % (folder, i)
-            files = os.listdir(dir)
-            if mini > len(files):
-                mini = len(files)
-
         for i in range(OUTPUT_NODES):
             dir = "%s/%d" % (folder, i)
             files = os.listdir(dir)