Merge pull request #921 from reyoung/feature/refine_demo_dataprovider

Feature/refine demo dataprovider

demo/gan/data/download_cifar.sh

@@ -1,3 +1,4 @@
+#!/bin/bash
 # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

demo/image_classification/data/download_cifar.sh

@@ -1,3 +1,4 @@
+#!/bin/bash
 # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

demo/image_classification/image_provider.py

@@ -21,7 +21,7 @@ from paddle.trainer.PyDataProvider2 import *
 
 #
 # {'img_size': 32,
-# 'settings': <paddle.trainer.PyDataProviderWrapper.Cls instance at 0x7fea27cb6050>,
+# 'settings': a global object,
 # 'color': True,
 # 'mean_img_size': 32,
 # 'meta': './data/cifar-out/batches/batches.meta',
@@ -50,10 +50,10 @@ def hook(settings, img_size, mean_img_size, num_classes, color, meta, use_jpeg,
 
     settings.logger.info('Image size: %s', settings.img_size)
     settings.logger.info('Meta path: %s', settings.meta_path)
-    settings.input_types = [
-        dense_vector(settings.img_raw_size),  # image feature
-        integer_value(settings.num_classes)
-    ]  # labels
+    settings.input_types = {
+        'image': dense_vector(settings.img_raw_size),
+        'label': integer_value(settings.num_classes)
+    }
 
     settings.logger.info('DataProvider Initialization finished')
 
@@ -83,4 +83,7 @@ def processData(settings, file_list):
                         img, settings.img_mean, settings.img_size,
                         settings.is_train, settings.color)
                     label = data['labels'][i]
-                    yield img_feat.astype('float32'), int(label)
+                    yield {
+                        'image': img_feat.astype('float32'),
+                        'label': int(label)
+                    }

demo/introduction/.gitignore

@@ -0,0 +1,5 @@
+dataprovider.pyc
+empty.list
+train.log
+output
+train.list

demo/introduction/dataprovider.py

@@ -17,8 +17,10 @@ import random
 
 
 # define data types of input: 2 real numbers
-@provider(input_types=[dense_vector(1), dense_vector(1)], use_seq=False)
+@provider(
+    input_types={'x': dense_vector(1),
+                 'y': dense_vector(1)}, use_seq=False)
 def process(settings, input_file):
     for i in xrange(2000):
         x = random.random()
-        yield [x], [2 * x + 0.3]
+        yield {'x': [x], 'y': [2 * x + 0.3]}

demo/introduction/trainer_config.py

@@ -15,11 +15,8 @@
 from paddle.trainer_config_helpers import *
 
 # 1. read data. Suppose you saved above python code as dataprovider.py
-data_file = 'empty.list'
-with open(data_file, 'w') as f:
-    f.writelines(' ')
 define_py_data_sources2(
-    train_list=data_file,
+    train_list=['no_matter.txt'],
     test_list=None,
     module='dataprovider',
     obj='process',

demo/quick_start/.gitignore

@@ -8,6 +8,8 @@ data/test.list
 data/test.txt
 data/train.list
 data/train.txt
+data/pred.list
+data/pred.txt
 dataprovider_copy_1.py
 train.log
 output

demo/quick_start/dataprovider_bow.py

@@ -31,16 +31,16 @@ def initializer(settings, dictionary, **kwargs):
 
     # setting.input_types specifies what the data types the data provider
     # generates.
-    settings.input_types = [
+    settings.input_types = {
         # The first input is a sparse_binary_vector,
         # which means each dimension of the vector is either 0 or 1. It is the
         # bag-of-words (BOW) representation of the texts.
-        sparse_binary_vector(len(dictionary)),
+        'word': sparse_binary_vector(len(dictionary)),
         # The second input is an integer. It represents the category id of the
         # sample. 2 means there are two labels in the dataset.
         # (1 for positive and 0 for negative)
-        integer_value(2)
-    ]
+        'label': integer_value(2)
+    }
 
 
 # Delaring a data provider. It has an initializer 'data_initialzer'.
@@ -67,12 +67,12 @@ def process(settings, file_name):
             # Return the features for the current comment. The first is a list
             # of ids representing a 0-1 binary sparse vector of the text,
             # the second is the integer id of the label.
-            yield word_vector, int(label)
+            yield {'word': word_vector, 'label': int(label)}
 
 
 def predict_initializer(settings, dictionary, **kwargs):
     settings.word_dict = dictionary
-    settings.input_types = [sparse_binary_vector(len(dictionary))]
+    settings.input_types = {'word': sparse_binary_vector(len(dictionary))}
 
 
 # Declaring a data provider for prediction. The difference with process
@@ -83,4 +83,4 @@ def process_predict(settings, file_name):
         for line in f:
             comment = line.strip().split()
             word_vector = [settings.word_dict.get(w, UNK_IDX) for w in comment]
-            yield word_vector
+            yield {'word': word_vector}

demo/quick_start/dataprovider_emb.py

@@ -19,13 +19,13 @@ UNK_IDX = 0
 
 def initializer(settings, dictionary, **kwargs):
     settings.word_dict = dictionary
-    settings.input_types = [
+    settings.input_types = {
         # Define the type of the first input as sequence of integer.
         # The value of the integers range from 0 to len(dictrionary)-1
-        integer_value_sequence(len(dictionary)),
+        'word': integer_value_sequence(len(dictionary)),
         # Define the second input for label id
-        integer_value(2)
-    ]
+        'label': integer_value(2)
+    }
 
 
 @provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM)
@@ -35,15 +35,12 @@ def process(settings, file_name):
             label, comment = line.strip().split('\t')
             words = comment.split()
             word_slot = [settings.word_dict.get(w, UNK_IDX) for w in words]
-            yield word_slot, int(label)
+            yield {'word': word_slot, 'label': int(label)}
 
 
 def predict_initializer(settings, dictionary, **kwargs):
     settings.word_dict = dictionary
-    settings.input_types = [
-        integer_value(
-            len(dictionary), seq_type=SequenceType.SEQUENCE)
-    ]
+    settings.input_types = {'word': integer_value_sequence(len(dictionary))}
 
 
 @provider(init_hook=predict_initializer, should_shuffle=False)
@@ -52,4 +49,4 @@ def process_predict(settings, file_name):
         for line in f:
             comment = line.strip().split()
             word_slot = [settings.word_dict.get(w, UNK_IDX) for w in comment]
-            yield word_slot
+            yield {'word': word_slot}

demo/recommendation/common_utils.py

@@ -17,13 +17,14 @@ from paddle.trainer.PyDataProvider2 import *
 def meta_to_header(meta, name):
     metas = meta[name]['__meta__']['raw_meta']
     for each_meta in metas:
+        slot_name = each_meta.get('name', '%s_id' % name)
         if each_meta['type'] == 'id':
-            yield integer_value(each_meta['max'])
+            yield slot_name, integer_value(each_meta['max'])
         elif each_meta['type'] == 'embedding':
             is_seq = each_meta['seq'] == 'sequence'
-            yield integer_value(
+            yield slot_name, integer_value(
                 len(each_meta['dict']),
                 seq_type=SequenceType.SEQUENCE
                 if is_seq else SequenceType.NO_SEQUENCE)
         elif each_meta['type'] == 'one_hot_dense':
-            yield dense_vector(len(each_meta['dict']))
+            yield slot_name, dense_vector(len(each_meta['dict']))

demo/recommendation/dataprovider.py

@@ -16,6 +16,14 @@ from paddle.trainer.PyDataProvider2 import *
 import common_utils  # parse
 
 
+def __list_to_map__(lst):
+    ret_val = dict()
+    for each in lst:
+        k, v = each
+        ret_val[k] = v
+    return ret_val
+
+
 def hook(settings, meta, **kwargs):
     """
     Init hook is invoked before process data. It will set obj.slots and store
@@ -34,12 +42,16 @@ def hook(settings, meta, **kwargs):
     #    second part is user features.
     #    final part is rating score.
     # header is a list of [USE_SEQ_OR_NOT?, SlotType]
-    headers = list(common_utils.meta_to_header(meta, 'movie'))
-    headers.extend(list(common_utils.meta_to_header(meta, 'user')))
-    headers.append(dense_vector(1))  # Score
+    movie_headers = list(common_utils.meta_to_header(meta, 'movie'))
+    settings.movie_names = [h[0] for h in movie_headers]
+    headers = movie_headers
+    user_headers = list(common_utils.meta_to_header(meta, 'user'))
+    settings.user_names = [h[0] for h in user_headers]
+    headers.extend(user_headers)
+    headers.append(("rating", dense_vector(1)))  # Score
 
     # slot types.
-    settings.input_types = headers
+    settings.input_types = __list_to_map__(headers)
     settings.meta = meta
 
 
@@ -57,20 +69,20 @@ def process(settings, filename):
             movie_meta = settings.meta['movie'][movie_id]
             user_meta = settings.meta['user'][user_id]
 
-            outputs = [movie_id - 1]
+            outputs = [('movie_id', movie_id - 1)]
 
             # Then add movie features
-            for each_meta in movie_meta:
-                outputs.append(each_meta)
+            for i, each_meta in enumerate(movie_meta):
+                outputs.append((settings.movie_names[i + 1], each_meta))
 
             # Then add user id.
-            outputs.append(user_id - 1)
+            outputs.append(('user_id', user_id - 1))
 
             # Then add user features.
-            for each_meta in user_meta:
-                outputs.append(each_meta)
+            for i, each_meta in enumerate(user_meta):
+                outputs.append((settings.user_names[i + 1], each_meta))
 
             # Finally, add score
-            outputs.append([score])
+            outputs.append(('rating', [score]))
             # Return data to paddle
-            yield outputs
+            yield __list_to_map__(outputs)

demo/recommendation/prediction.py

@@ -34,8 +34,8 @@ if __name__ == '__main__':
     network.loadParameters(model_path)
     with open('./data/meta.bin', 'rb') as f:
         meta = pickle.load(f)
-        headers = list(meta_to_header(meta, 'movie'))
-        headers.extend(list(meta_to_header(meta, 'user')))
+        headers = [h[1] for h in meta_to_header(meta, 'movie')]
+        headers.extend([h[1] for h in meta_to_header(meta, 'user')])
         cvt = DataProviderConverter(headers)
         while True:
             movie_id = int(raw_input("Input movie_id: "))

demo/recommendation/preprocess.sh

@@ -25,7 +25,7 @@ python meta_generator.py $dir meta.bin --config=meta_config.json
 echo 'split train/test file'
 python split.py $dir/ratings.dat --delimiter=${delimiter} --test_ratio=0.1
 echo 'shuffle train file'
-shuf $dir/ratings.dat.train > ratings.dat.train
+gshuf $dir/ratings.dat.train > ratings.dat.train
 cp $dir/ratings.dat.test .
 echo "./data/ratings.dat.train" > train.list
 echo "./data/ratings.dat.test" > test.list

demo/semantic_role_labeling/.gitignore

@@ -8,3 +8,7 @@ data/test.wsj.seq_pair
 data/test.wsj.words
 data/tgt.dict
 output
+data/emb
+data/targetDict.txt
+data/verbDict.txt
+data/wordDict.txt

python/paddle/trainer_config_helpers/data_sources.py

@@ -69,7 +69,7 @@ def define_py_data_source(file_list,
     """
     if isinstance(file_list, list):
         file_list_name = 'train.list'
-        if isinstance(cls, TestData):
+        if cls == TestData:
             file_list_name = 'test.list'
         with open(file_list_name, 'w') as f:
             f.writelines(file_list)

python/paddle/trainer_config_helpers/layers.py

@@ -970,7 +970,7 @@ def pooling_layer(input,
     :param layer_attr: The Extra Attributes for layer, such as dropout.
     :type layer_attr: ExtraLayerAttribute|None
     :return: LayerOutput object.
-    :rtype: LayerType
+    :rtype: LayerOutput
     """
     extra_dict = dict()
     # noinspection PyUnresolvedReferences