Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
S
seminar-breakout
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Issues
0
Issues
0
List
Boards
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Analytics
Analytics
CI / CD
Repository
Value Stream
Wiki
Wiki
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
Shashank Suhas
seminar-breakout
Commits
9b318943
Commit
9b318943
authored
May 23, 2019
by
Yuxin Wu
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
grow lmdb map_size (fix #1209)
parent
413059b1
Changes
2
Show whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
27 additions
and
4 deletions
+27
-4
examples/keras/mnist-keras.py
examples/keras/mnist-keras.py
+2
-1
tensorpack/dataflow/serialize.py
tensorpack/dataflow/serialize.py
+25
-3
No files found.
examples/keras/mnist-keras.py
View file @
9b318943
...
@@ -20,7 +20,8 @@ KL = keras.layers
...
@@ -20,7 +20,8 @@ KL = keras.layers
This is an mnist example demonstrating how to use Keras symbolic function inside tensorpack.
This is an mnist example demonstrating how to use Keras symbolic function inside tensorpack.
This way you can define models in Keras-style, and benefit from the more efficeint trainers in tensorpack.
This way you can define models in Keras-style, and benefit from the more efficeint trainers in tensorpack.
Note: this example does not work for replicated-style data-parallel trainers.
Note: this example does not work for replicated-style data-parallel trainers, so may be less efficient
for some models.
"""
"""
IMAGE_SIZE
=
28
IMAGE_SIZE
=
28
...
...
tensorpack/dataflow/serialize.py
View file @
9b318943
...
@@ -3,6 +3,7 @@
...
@@ -3,6 +3,7 @@
import
numpy
as
np
import
numpy
as
np
import
os
import
os
import
platform
from
collections
import
defaultdict
from
collections
import
defaultdict
from
..utils
import
logger
from
..utils
import
logger
...
@@ -47,10 +48,31 @@ class LMDBSerializer():
...
@@ -47,10 +48,31 @@ class LMDBSerializer():
assert
not
os
.
path
.
isfile
(
os
.
path
.
join
(
path
,
'data.mdb'
)),
"LMDB file exists!"
assert
not
os
.
path
.
isfile
(
os
.
path
.
join
(
path
,
'data.mdb'
)),
"LMDB file exists!"
else
:
else
:
assert
not
os
.
path
.
isfile
(
path
),
"LMDB file {} exists!"
.
format
(
path
)
assert
not
os
.
path
.
isfile
(
path
),
"LMDB file {} exists!"
.
format
(
path
)
# It's OK to use super large map_size on Linux, but not on other platforms
# See: https://github.com/NVIDIA/DIGITS/issues/206
map_size
=
1099511627776
*
2
if
platform
.
system
()
==
'Linux'
else
128
*
10
**
6
db
=
lmdb
.
open
(
path
,
subdir
=
isdir
,
db
=
lmdb
.
open
(
path
,
subdir
=
isdir
,
map_size
=
1099511627776
*
2
,
readonly
=
False
,
map_size
=
map_size
,
readonly
=
False
,
meminit
=
False
,
map_async
=
True
)
# need sync() at the end
meminit
=
False
,
map_async
=
True
)
# need sync() at the end
size
=
_reset_df_and_get_size
(
df
)
size
=
_reset_df_and_get_size
(
df
)
# put data into lmdb, and doubling the size if full.
# Ref: https://github.com/NVIDIA/DIGITS/pull/209/files
def
put_or_grow
(
txn
,
key
,
value
):
try
:
txn
.
put
(
key
,
value
)
return
txn
except
lmdb
.
MapFullError
:
pass
txn
.
abort
()
curr_size
=
db
.
info
()[
'map_size'
]
new_size
=
curr_size
*
2
logger
.
info
(
"Doubling LMDB map_size to {:.2f}GB"
.
format
(
new_size
/
10
**
9
))
db
.
set_mapsize
(
new_size
)
txn
=
db
.
begin
(
write
=
True
)
txn
=
put_or_grow
(
txn
,
key
,
value
)
return
txn
with
get_tqdm
(
total
=
size
)
as
pbar
:
with
get_tqdm
(
total
=
size
)
as
pbar
:
idx
=
-
1
idx
=
-
1
...
@@ -58,7 +80,7 @@ class LMDBSerializer():
...
@@ -58,7 +80,7 @@ class LMDBSerializer():
# although it has a context manager interface
# although it has a context manager interface
txn
=
db
.
begin
(
write
=
True
)
txn
=
db
.
begin
(
write
=
True
)
for
idx
,
dp
in
enumerate
(
df
):
for
idx
,
dp
in
enumerate
(
df
):
txn
.
put
(
u'{:08}'
.
format
(
idx
)
.
encode
(
'ascii'
),
dumps
(
dp
))
txn
=
put_or_grow
(
txn
,
u'{:08}'
.
format
(
idx
)
.
encode
(
'ascii'
),
dumps
(
dp
))
pbar
.
update
()
pbar
.
update
()
if
(
idx
+
1
)
%
write_frequency
==
0
:
if
(
idx
+
1
)
%
write_frequency
==
0
:
txn
.
commit
()
txn
.
commit
()
...
@@ -67,7 +89,7 @@ class LMDBSerializer():
...
@@ -67,7 +89,7 @@ class LMDBSerializer():
keys
=
[
u'{:08}'
.
format
(
k
)
.
encode
(
'ascii'
)
for
k
in
range
(
idx
+
1
)]
keys
=
[
u'{:08}'
.
format
(
k
)
.
encode
(
'ascii'
)
for
k
in
range
(
idx
+
1
)]
with
db
.
begin
(
write
=
True
)
as
txn
:
with
db
.
begin
(
write
=
True
)
as
txn
:
txn
.
put
(
b
'__keys__'
,
dumps
(
keys
))
txn
=
put_or_grow
(
txn
,
b
'__keys__'
,
dumps
(
keys
))
logger
.
info
(
"Flushing database ..."
)
logger
.
info
(
"Flushing database ..."
)
db
.
sync
()
db
.
sync
()
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment