the code is not clean enough yet

a very original version

Author: cyente

NGNN/Config.py

@@ -0,0 +1,174 @@
+"""
+find which config to use and select
+
+"""
+
+
+def Config(reader, flag="azenuz_small"):
+    if flag == "azenuz_small":
+        return configazenuz_small(reader)
+    elif flag == "small_amazon":
+        return smallconfig_amazon(reader)
+    elif flag == "test_ptb":
+        return testconfig_ptb(reader)
+    elif flag == "small_amazontree":
+        return smallconfig_amazontree(reader)
+    elif flag == "small_amazontree1":
+        return smallconfig_amazontree1(reader)
+    elif flag == "small_amazontree2":
+        return smallconfig_amazontree2(reader)
+    elif flag == "small_amazontree3":
+        return smallconfig_amazontree3(reader)
+    else:
+        raise ValueError("Invalid model: %s", flag)
+
+
+class configazenuz_small(object):
+    epoch_num = 1000
+    train_batch_size = 128  # 128
+    train_step_size = 4 # 20
+    valid_batch_size = 128  # 20
+    valid_step_size = 4
+    hidden_size = 16  # 512
+
+    lstm_forget_bias = 0.0
+    # max_grad_norm = 0.25
+    max_grad_norm = 1
+    init_scale = 0.05
+    learning_rate = 0.01  # 0.001  # 0.2
+    decay = 0.5
+    decay_when = 0.002  # AUC
+    decay_epoch = 200
+    sgd_opt = 'RMSProp'
+    beta = 0.0001
+    GNN_step = 3
+    dropout_prob = 0
+    adagrad_eps = 1e-5
+    gpu = 0
+
+    def __init__(self, loader):
+        self.node_num = len(loader.train_set[0]) # the num of features
+        self.feature_num = loader.feature_num
+        assert self.node_num == len(self.feature_num)
+
+        self.user_size = len(loader.train_set)
+        max_step = 0
+        for line in loader.train_set:
+            if max_step < len(line):
+                max_step = len(line)
+        self.maxstep_size = max_step + 1
+
+        print "usernum", self.user_size
+        print 'node_num', self.node_num
+        print 'feature_num', self.feature_num
+        print "maxstep_size %d" % self.maxstep_size
+        print "gpu_id {}".format(self.gpu)
+        print "learning_rate {}".format(self.learning_rate)
+
+
+class smallconfig_amazon(object):
+    epoch_num = 1000
+    train_batch_size = 1  # 128
+    train_step_size = 4 # 20
+    valid_batch_size = 1  # 128
+    valid_step_size = 4  # 20
+    test_batch_size = 1  # 20
+    test_step_size = 4
+
+    def __init__(self, loader):
+        vec = loader.itemdict.values()
+        # print vec
+        vec_r, vec_c = zip(*vec)
+        self.vocab_size = (max(vec_r) + 2, max(vec_c) + 2)
+        # self.vocab_size = loader.num_items  # 10000
+        max_step = 0
+        for line in loader.train_set:
+            if max_step < len(line):
+                max_step = len(line)
+        self.maxstep_size = max_step + 1
+        print "word-embedding %d" % self.word_embedding_dim
+
+
+class smallconfig_amazontree(object):
+    epoch_num = 1000
+    train_batch_size = 100  # 128
+    train_step_size = 4 # 20
+    valid_batch_size = 100  # 128
+    valid_step_size = 4  # 20
+    test_batch_size = 100  # 20
+    test_step_size = 4
+    word_embedding_dim = 100  # 512
+    lstm_layers = 1
+    lstm_size = 100  # 512
+    lstm_forget_bias = 0.0
+    # max_grad_norm = 0.25
+    max_grad_norm = 1
+    init_scale = 0.05
+    learning_rate = 1  # 0.2
+    decay = 0.5
+    decay_when = 0.002  # AUC
+    dropout_prob = 0
+    adagrad_eps = 1e-5
+    gpu = 1
+
+    def __init__(self, loader):
+        vec = loader.itemdict.values()
+        # vec_r, vec_c = zip(*vec)
+        self.tree_size = len(zip(*vec)) - 1
+        cat = [max(voc) + 2 for voc in zip(*vec)]
+        self.vocab_size = tuple(cat)
+        # self.vocab_size = loader.num_items  # 10000
+        max_step = 0
+        self.loader = loader
+        for line in loader.train_set:
+            if max_step < len(line):
+                max_step = len(line)
+        self.user_size = len(loader.train_set)
+        self.maxstep_size = max_step + 1
+        self.layer_embed = (0.2, 0.3, 0.3, 0.2)
+        self.vocab_size_all = len(loader.itemdict)
+        assert len(self.layer_embed) == self.tree_size
+        print "usernum", self.user_size
+        print 'itemnum_vocab_size_all', self.vocab_size_all
+        print 'itemnum_vocab_size', self.vocab_size
+        print "word-embedding %d" % self.word_embedding_dim
+
+
+class smallconfig_amazontree1(smallconfig_amazontree):
+    def __init__(self, loader):
+        smallconfig_amazontree.__init__(self, loader)
+        self.layer_embed = (0.1, 0.1, 0.3, 0.5)
+        # self.word_embedding_dim = (self.word_embedding_dim / self.layer_embed[-1]) * sum(self.layer_embed)
+
+
+class smallconfig_amazontree2(smallconfig_amazontree):
+    def __init__(self, loader):
+        smallconfig_amazontree.__init__(self, loader)
+        self.layer_embed = (0, 0, 0, 1)
+        # self.word_embedding_dim = (self.word_embedding_dim / self.layer_embed[-1]) * sum(self.layer_embed)
+
+
+class smallconfig_amazontree3(smallconfig_amazontree):
+    def __init__(self, loader):
+        smallconfig_amazontree.__init__(self, loader)
+        self.layer_embed = (0.6, 0.1, 0.1, 0.2)
+        # self.word_embedding_dim = (self.word_embedding_dim / self.layer_embed[-1]) * sum(self.layer_embed)
+
+
+
+class testconfig_ptb(object):
+      """Tiny config, for testing."""
+      init_scale = 0.1
+      learning_rate = 1.0
+      max_grad_norm = 1
+      num_layers = 1
+      num_steps = 2
+      hidden_size = 2
+      max_epoch = 1
+      max_max_epoch = 1
+      keep_prob = 1.0
+      lr_decay = 0.5
+      batch_size = 20
+
+      def __init__(self, reader):
+          self.vocab_size = len(reader.vocab.words)  # 10000