Skip to content

Commit 554cea6

Browse files
seiriosPlusseiriosPlus
committed
move most codes to uci_housing #9660 
1 parent 3a3ff62 commit 554cea6

File tree

3 files changed

+80
-171
lines changed

3 files changed

+80
-171
lines changed

‎doc/fluid/getstarted/quickstart_cn.rst

Lines changed: 18 additions & 81 deletions
Original file line numberDiff line numberDiff line change
@@ -25,94 +25,31 @@ PaddlePaddle支持使用pip快速安装，目前支持CentOS 6以上, Ubuntu 14.
2525
创建一个 housing.py 并粘贴此Python代码：
2626

2727
.. code-block:: python
28-
29-
import sys
30-
31-
import math
32-
import numpy
33-
34-
import paddle.fluid as fluid
35-
import paddle.fluid.core as core
3628
import paddle
29+
import paddle.fluid as fluid
3730
38-
def train(save_dirname):
39-
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
40-
y_predict = fluid.layers.fc(input=x, size=1, act=None)
41-
y = fluid.layers.data(name='y', shape=[1], dtype='float32')
42-
43-
cost = fluid.layers.square_error_cost(input=y_predict, label=y)
44-
avg_cost = fluid.layers.mean(cost)
45-
46-
sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
47-
optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost)
48-
49-
BATCH_SIZE = 20
50-
51-
train_reader = paddle.batch(
52-
paddle.reader.shuffle(paddle.dataset.uci_housing.train(), buf_size=500), batch_size=BATCH_SIZE)
53-
54-
place = fluid.CPUPlace()
55-
exe = fluid.Executor(place)
56-
57-
feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
58-
exe.run(fluid.default_startup_program())
59-
60-
main_program = fluid.default_main_program()
6131
62-
PASS_NUM = 100
63-
for pass_id in range(PASS_NUM):
64-
for data in train_reader():
65-
avg_loss_value, = exe.run(main_program,
66-
feed=feeder.feed(data),
67-
fetch_list=[avg_cost])
68-
if avg_loss_value[0] < 10.0:
69-
if save_dirname is not None:
70-
fluid.io.save_inference_model(save_dirname, ['x'],
71-
[y_predict], exe)
72-
return
73-
if math.isnan(float(avg_loss_value)):
74-
sys.exit("got NaN loss, training failed.")
75-
raise AssertionError("Fit a line cost is too large, {0:2.2}".format(
76-
avg_loss_value[0]))
32+
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
33+
place = fluid.CPUPlace()
34+
exe = fluid.Executor(place=place)
35+
feeder = fluid.DataFeeder(place=place, feed_list=[x])
7736
78-
def infer(save_dirname):
79-
place = fluid.CPUPlace()
80-
exe = fluid.Executor(place)
37+
with fluid.scope_guard(fluid.core.Scope()):
38+
parameter_model = paddle.dataset.uci_housing.fluid_model()
8139
82-
probs = []
40+
[inference_program, feed_target_names,fetch_targets] = \
41+
fluid.io.load_inference_model(parameter_model, exe)
8342
84-
inference_scope = fluid.core.Scope()
85-
with fluid.scope_guard(inference_scope):
86-
# Use fluid.io.load_inference_model to obtain the inference program desc,
87-
# the feed_target_names (the names of variables that will be feeded
88-
# data using feed operators), and the fetch_targets (variables that
89-
# we want to obtain data from using fetch operators).
90-
[inference_program, feed_target_names,
91-
fetch_targets] = fluid.io.load_inference_model(save_dirname, exe)
43+
predict_reader = paddle.batch(paddle.dataset.uci_housing.predict_reader(), batch_size=20)
9244
93-
# The input's dimension should be 2-D and the second dim is 13
94-
# The input data should be >= 0
95-
batch_size = 10
96-
tensor_x = numpy.random.uniform(0, 10,
97-
[batch_size, 13]).astype("float32")
98-
assert feed_target_names[0] == 'x'
99-
results = exe.run(inference_program,
100-
feed={feed_target_names[0]: tensor_x},
45+
results = []
46+
for data in predict_reader():
47+
result = exe.run(inference_program,
48+
feed=feeder.feed(data),
10149
fetch_list=fetch_targets)
102-
probs.append(results)
50+
results.append(result)
10351
104-
for i in xrange(len(probs)):
105-
print(probs[i][0] * 1000)
106-
print('Predicted price: ${0}'.format(probs[i][0] * 1000))
107-
108-
def main():
109-
# Directory for saving the trained model
110-
save_dirname = "fit_a_line.inference.model"
111-
112-
train(save_dirname)
113-
infer(save_dirname)
114-
115-
if __name__=="__main__":
116-
main()
117-
52+
for res in results:
53+
for i in xrange(len(res[0])):
54+
print 'Predicted price: ${:,.2f}'.format(res[0][i][0] * 1000)
11855
执行 :code:`python housing.py` 瞧！ 它应该打印出预测住房数据的清单。

‎doc/fluid/getstarted/quickstart_en.rst

Lines changed: 28 additions & 88 deletions
Original file line numberDiff line numberDiff line change
@@ -28,93 +28,33 @@ code:
2828

2929

3030
.. code-block:: python
31-
import sys
32-
33-
import math
34-
import numpy
35-
36-
import paddle.fluid as fluid
37-
import paddle.fluid.core as core
38-
import paddle
39-
40-
def train(save_dirname):
41-
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
42-
y_predict = fluid.layers.fc(input=x, size=1, act=None)
43-
y = fluid.layers.data(name='y', shape=[1], dtype='float32')
44-
45-
cost = fluid.layers.square_error_cost(input=y_predict, label=y)
46-
avg_cost = fluid.layers.mean(cost)
47-
48-
sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
49-
optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost)
50-
51-
BATCH_SIZE = 20
52-
53-
train_reader = paddle.batch(
54-
paddle.reader.shuffle(paddle.dataset.uci_housing.train(), buf_size=500), batch_size=BATCH_SIZE)
55-
56-
place = fluid.CPUPlace()
57-
exe = fluid.Executor(place)
58-
59-
feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
60-
exe.run(fluid.default_startup_program())
61-
62-
main_program = fluid.default_main_program()
63-
64-
PASS_NUM = 100
65-
for pass_id in range(PASS_NUM):
66-
for data in train_reader():
67-
avg_loss_value, = exe.run(main_program,
68-
feed=feeder.feed(data),
69-
fetch_list=[avg_cost])
70-
if avg_loss_value[0] < 10.0:
71-
if save_dirname is not None:
72-
fluid.io.save_inference_model(save_dirname, ['x'],
73-
[y_predict], exe)
74-
return
75-
if math.isnan(float(avg_loss_value)):
76-
sys.exit("got NaN loss, training failed.")
77-
raise AssertionError("Fit a line cost is too large, {0:2.2}".format(
78-
avg_loss_value[0]))
79-
80-
def infer(save_dirname):
81-
place = fluid.CPUPlace()
82-
exe = fluid.Executor(place)
83-
84-
probs = []
85-
86-
inference_scope = fluid.core.Scope()
87-
with fluid.scope_guard(inference_scope):
88-
# Use fluid.io.load_inference_model to obtain the inference program desc,
89-
# the feed_target_names (the names of variables that will be feeded
90-
# data using feed operators), and the fetch_targets (variables that
91-
# we want to obtain data from using fetch operators).
92-
[inference_program, feed_target_names,
93-
fetch_targets] = fluid.io.load_inference_model(save_dirname, exe)
94-
95-
# The input's dimension should be 2-D and the second dim is 13
96-
# The input data should be >= 0
97-
batch_size = 10
98-
tensor_x = numpy.random.uniform(0, 10,
99-
[batch_size, 13]).astype("float32")
100-
assert feed_target_names[0] == 'x'
101-
results = exe.run(inference_program,
102-
feed={feed_target_names[0]: tensor_x},
103-
fetch_list=fetch_targets)
104-
probs.append(results)
105-
106-
for i in xrange(len(probs)):
107-
print(probs[i][0] * 1000)
108-
print('Predicted price: ${0}'.format(probs[i][0] * 1000))
109-
110-
def main():
111-
# Directory for saving the trained model
112-
save_dirname = "fit_a_line.inference.model"
113-
114-
train(save_dirname)
115-
infer(save_dirname)
116-
117-
if __name__=="__main__":
118-
main()
31+
import paddle
32+
import paddle.fluid as fluid
33+
34+
35+
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
36+
place = fluid.CPUPlace()
37+
exe = fluid.Executor(place=place)
38+
feeder = fluid.DataFeeder(place=place, feed_list=[x])
39+
40+
with fluid.scope_guard(fluid.core.Scope()):
41+
parameter_model = paddle.dataset.uci_housing.fluid_model()
42+
43+
[inference_program, feed_target_names,fetch_targets] = \
44+
fluid.io.load_inference_model(parameter_model, exe)
45+
46+
predict_reader = paddle.batch(paddle.dataset.uci_housing.predict_reader(), batch_size=20)
47+
48+
results = []
49+
for data in predict_reader():
50+
result = exe.run(inference_program,
51+
feed=feeder.feed(data),
52+
fetch_list=fetch_targets)
53+
results.append(result)
54+
55+
for res in results:
56+
for i in xrange(len(res[0])):
57+
print 'Predicted price: ${:,.2f}'.format(res[0][i][0] * 1000)
58+
11959
Run :code:`python housing.py` and voila! It should print out a list of predictions
12060
for the test housing data.

‎python/paddle/dataset/uci_housing.py

Lines changed: 34 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,11 @@
1919
parse training set and test set into paddle reader creators.
2020
"""
2121

22+
import os
23+
2224
import numpy as np
25+
import tempfile
26+
import tarfile
2327
import os
2428
import paddle.dataset.common
2529

@@ -34,8 +38,9 @@
3438

3539
UCI_TRAIN_DATA = None
3640
UCI_TEST_DATA = None
37-
URL_MODEL = 'https://github.com/PaddlePaddle/book/raw/develop/01.fit_a_line/fit_a_line.tar'
38-
MD5_MODEL = '52fc3da8ef3937822fcdd87ee05c0c9b'
41+
42+
FLUID_URL_MODEL = 'https://github.com/PaddlePaddle/book/raw/develop/01.fit_a_line/fluid/fit_a_line.fluid.tar'
43+
FLUID_MD5_MODEL = '6e6dd637ccd5993961f68bfbde46090b'
3944

4045

4146
def feature_range(maximums, minimums):
@@ -112,6 +117,33 @@ def reader():
112117

113118
return reader
114119

120+
def fluid_model():
121+
parameter_tar = paddle.dataset.common.download(FLUID_URL_MODEL, 'uci_housing', FLUID_MD5_MODEL, 'fit_a_line.fluid.tar')
122+
123+
tar = tarfile.TarFile(parameter_tar, mode='r')
124+
dirpath = tempfile.mkdtemp()
125+
tar.extractall(path=dirpath)
126+
127+
return dirpath
128+
129+
def predict_reader():
130+
"""
131+
UCI_HOUSING test set creator.
132+
133+
It returns a reader creator, each sample in the reader is features after
134+
normalization and price number.
135+
136+
:return: Test reader creator
137+
:rtype: callable
138+
"""
139+
global UCI_TEST_DATA
140+
load_data(paddle.dataset.common.download(URL, 'uci_housing', MD5))
141+
142+
def reader():
143+
for d in UCI_TEST_DATA:
144+
yield (d[:-1],)
145+
146+
return reader
115147

116148
def fetch():
117149
paddle.dataset.common.download(URL, 'uci_housing', MD5)

0 commit comments

Comments
 (0)