From ea25eef375c33d01e2f28f875dbcb5650596466d Mon Sep 17 00:00:00 2001
From: qiaolongfei <qiaolongfei@baidu.com>
Date: Sun, 16 Apr 2017 16:16:17 +0800
Subject: [PATCH] word2vec demo support sparse remote update

---
 .../word2vec/{train_v2.py => api_train_v2.py} | 28 +++++++++++++++----
 paddle/api/PaddleAPI.h                        |  2 +-
 paddle/api/ParameterUpdater.cpp               |  4 +--
 3 files changed, 25 insertions(+), 9 deletions(-)
 rename demo/word2vec/{train_v2.py => api_train_v2.py} (80%)

diff --git a/demo/word2vec/train_v2.py b/demo/word2vec/api_train_v2.py
similarity index 80%
rename from demo/word2vec/train_v2.py
rename to demo/word2vec/api_train_v2.py
index 7d952b446f9..eb61a7250fb 100644
--- a/demo/word2vec/train_v2.py
+++ b/demo/word2vec/api_train_v2.py
@@ -2,26 +2,38 @@ import math
 
 import paddle.v2 as paddle
 
-dictsize = 1953
 embsize = 32
 hiddensize = 256
 N = 5
 
 
 def wordemb(inlayer):
-    wordemb = paddle.layer.table_projection(
+    wordemb = paddle.layer.embedding(
         input=inlayer,
         size=embsize,
         param_attr=paddle.attr.Param(
             name="_proj",
             initial_std=0.001,
             learning_rate=1,
-            l2_rate=0, ))
+            l2_rate=0,
+            sparse_update=True))
     return wordemb
 
 
 def main():
-    paddle.init(use_gpu=False, trainer_count=1)
+    # for local training
+    cluster_train = False
+
+    if not cluster_train:
+        paddle.init(use_gpu=False, trainer_count=1)
+    else:
+        paddle.init(
+            use_gpu=False,
+            trainer_count=1,
+            port=7164,
+            ports_num=1,
+            ports_num_for_sparse=1,
+            num_gradient_servers=1)
     word_dict = paddle.dataset.imikolov.build_dict()
     dict_size = len(word_dict)
     firstword = paddle.layer.data(
@@ -65,11 +77,15 @@ def main():
                     result.metrics)
 
     cost = paddle.layer.classification_cost(input=predictword, label=nextword)
+
     parameters = paddle.parameters.create(cost)
-    adam_optimizer = paddle.optimizer.Adam(
+    adagrad = paddle.optimizer.AdaGrad(
         learning_rate=3e-3,
         regularization=paddle.optimizer.L2Regularization(8e-4))
-    trainer = paddle.trainer.SGD(cost, parameters, adam_optimizer)
+    trainer = paddle.trainer.SGD(cost,
+                                 parameters,
+                                 adagrad,
+                                 is_local=not cluster_train)
     trainer.train(
         paddle.batch(paddle.dataset.imikolov.train(word_dict, N), 32),
         num_passes=30,
diff --git a/paddle/api/PaddleAPI.h b/paddle/api/PaddleAPI.h
index be6be556a73..d5120401217 100644
--- a/paddle/api/PaddleAPI.h
+++ b/paddle/api/PaddleAPI.h
@@ -821,7 +821,7 @@ public:
   static ParameterUpdater* createLocalUpdater(OptimizationConfig* config);
   static ParameterUpdater* createRemoteUpdater(OptimizationConfig* config,
                                                int passCount,
-                                               bool userSparseUpdater);
+                                               bool useSparseUpdater);
   ~ParameterUpdater();
 
   /**
diff --git a/paddle/api/ParameterUpdater.cpp b/paddle/api/ParameterUpdater.cpp
index ce2ac33d449..9dfd12ccbe7 100644
--- a/paddle/api/ParameterUpdater.cpp
+++ b/paddle/api/ParameterUpdater.cpp
@@ -29,11 +29,11 @@ ParameterUpdater *ParameterUpdater::createLocalUpdater(
 }
 
 ParameterUpdater *ParameterUpdater::createRemoteUpdater(
-    OptimizationConfig *config, int passCount, bool userSparseUpdater) {
+    OptimizationConfig *config, int passCount, bool useSparseUpdater) {
   auto updater = new ParameterUpdater();
   auto remoteUpdater = new paddle::RemoteParameterUpdater(
       config->m->getConfig(), passCount, nullptr);
-  if (userSparseUpdater) {
+  if (useSparseUpdater) {
     std::unique_ptr<paddle::ParameterUpdater> remoteUpdaterPtr;
     remoteUpdaterPtr.reset(remoteUpdater);
     auto sparseRemoteUpdater =
-- 
GitLab