From eab208a9c53404c88b5031275283565850224b5a Mon Sep 17 00:00:00 2001
From: zhengya01 <zhengya01@baidu.com>
Date: Thu, 28 Feb 2019 07:23:08 +0000
Subject: [PATCH] add multiview_simnet ce

---
 fluid/PaddleRec/multiview_simnet/.run_ce.sh | 11 ++++
 fluid/PaddleRec/multiview_simnet/_ce.py     | 58 +++++++++++++++++++++
 fluid/PaddleRec/multiview_simnet/train.py   | 36 +++++++++++++
 3 files changed, 105 insertions(+)
 create mode 100755 fluid/PaddleRec/multiview_simnet/.run_ce.sh
 create mode 100644 fluid/PaddleRec/multiview_simnet/_ce.py

diff --git a/fluid/PaddleRec/multiview_simnet/.run_ce.sh b/fluid/PaddleRec/multiview_simnet/.run_ce.sh
new file mode 100755
index 00000000..4aaed27e
--- /dev/null
+++ b/fluid/PaddleRec/multiview_simnet/.run_ce.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+export MKL_NUM_THREADS=1
+export OMP_NUM_THREADS=1
+
+
+export CPU_NUM=1
+export NUM_THREADS=1
+
+FLAGS_benchmark=true  python train.py --enable_ce | python _ce.py
+
diff --git a/fluid/PaddleRec/multiview_simnet/_ce.py b/fluid/PaddleRec/multiview_simnet/_ce.py
new file mode 100644
index 00000000..ce6cf7cf
--- /dev/null
+++ b/fluid/PaddleRec/multiview_simnet/_ce.py
@@ -0,0 +1,58 @@
+# this file is only used for continuous evaluation test!
+
+import os
+import sys
+sys.path.append(os.environ['ceroot'])
+from kpi import CostKpi
+from kpi import DurationKpi
+from kpi import AccKpi
+
+
+each_pass_duration_cpu1_thread1_kpi = DurationKpi('each_pass_duration_cpu1_thread1', 0.08, 0, actived=True)
+train_loss_cpu1_thread1_kpi = CostKpi('train_loss_cpu1_thread1', 0.08, 0)
+
+tracking_kpis = [
+        each_pass_duration_cpu1_thread1_kpi,
+        train_loss_cpu1_thread1_kpi,
+        ]
+
+
+def parse_log(log):
+    '''
+    This method should be implemented by model developers.
+
+    The suggestion:
+
+    each line in the log should be key, value, for example:
+
+    "
+    train_cost\t1.0
+    test_cost\t1.0
+    train_cost\t1.0
+    train_cost\t1.0
+    train_acc\t1.2
+    "
+    '''
+    for line in log.split('\n'):
+        fs = line.strip().split('\t')
+        print(fs)
+        if len(fs) == 3 and fs[0] == 'kpis':
+            kpi_name = fs[1]
+            kpi_value = float(fs[2])
+            yield kpi_name, kpi_value
+
+
+def log_to_ce(log):
+    kpi_tracker = {}
+    for kpi in tracking_kpis:
+        kpi_tracker[kpi.name] = kpi
+
+    for (kpi_name, kpi_value) in parse_log(log):
+        print(kpi_name, kpi_value)
+        kpi_tracker[kpi_name].add_record(kpi_value)
+        kpi_tracker[kpi_name].persist()
+
+
+if __name__ == '__main__':
+    log = sys.stdin.read()
+    log_to_ce(log)
diff --git a/fluid/PaddleRec/multiview_simnet/train.py b/fluid/PaddleRec/multiview_simnet/train.py
index b4a566d3..f098fd10 100644
--- a/fluid/PaddleRec/multiview_simnet/train.py
+++ b/fluid/PaddleRec/multiview_simnet/train.py
@@ -81,10 +81,19 @@ def parse_args():
         "for index processing")
     parser.add_argument(
         "--hidden_size", type=int, default=128, help="Hidden dim")
+    parser.add_argument(
+        '--enable_ce',
+        action='store_true',
+        help='If set, run the task with continuous evaluation logs.')
     return parser.parse_args()
 
 
 def start_train(args):
+    if args.enable_ce:
+        SEED = 102
+        fluid.default_startup_program().random_seed = SEED
+        fluid.default_startup_program().random_seed = SEED
+
     dataset = reader.SyntheticDataset(args.sparse_feature_dim, args.query_slots,
                                       args.title_slots)
     train_reader = paddle.batch(
@@ -115,7 +124,10 @@ def start_train(args):
     exe = fluid.Executor(place)
     exe.run(startup_program)
 
+    total_time = 0
+    ce_info = []
     for pass_id in range(args.epochs):
+        start_time = time.time()
         for batch_id, data in enumerate(train_reader()):
             loss_val, correct_val = exe.run(loop_program,
                                             feed=feeder.feed(data),
@@ -123,10 +135,34 @@ def start_train(args):
             logger.info("TRAIN --> pass: {} batch_id: {} avg_cost: {}, acc: {}"
                         .format(pass_id, batch_id, loss_val,
                                 float(correct_val) / args.batch_size))
+            ce_info.append(loss_val[0])
+        end_time = time.time()
+        total_time += end_time - start_time
         fluid.io.save_inference_model(args.model_output_dir,
                                       [val.name for val in all_slots],
                                       [avg_cost, correct], exe)
 
+    # only for ce
+    if args.enable_ce:
+        threads_num, cpu_num = get_cards(args)
+        epoch_idx = args.epochs 
+        ce_loss = 0
+        try:
+            ce_loss = ce_info[-2]
+        except:
+            logger.error("ce info error")
+
+        print("kpis\teach_pass_duration_cpu%s_thread%s\t%s" %
+                (cpu_num, threads_num, total_time / epoch_idx))
+        print("kpis\ttrain_loss_cpu%s_thread%s\t%s" %
+                (cpu_num, threads_num, ce_loss))
+
+
+def get_cards(args):
+    threads_num = os.environ.get('NUM_THREADS', 1)
+    cpu_num = os.environ.get('CPU_NUM', 1)
+    return int(threads_num), int(cpu_num)
+
 
 def main():
     args = parse_args()
-- 
GitLab