I am using below command to start the training of deepspeech model
%cd /content/DeepSpeech
!python3 DeepSpeech.py \
--drop_source_layers 2 --scorer /content/DeepSpeech/data/lm/kenlm-nigerian.scorer\
--train_cudnn True --early_stop True --es_epochs 6 --n_hidden 2048 --epochs 5 \
--export_dir /content/models/ --checkpoint_dir /content/model_checkpoints/ \
--train_files /content/train.csv --dev_files /content/dev.csv --test_files /content/test.csv \
--learning_rate 0.0001 --train_batch_size 64 --test_batch_size 32 --dev_batch_size 32 --export_file_name 'he_model_5' \
--max_to_keep 3
I keep getting following error again and again.
(0) Invalid argument: Not enough time for target transition sequence (required: 28, available: 24)0You can turn this error into a warning by using the flag ignore_longer_outputs_than_inputs
(1) Invalid argument: Not enough time for target transition sequence (required: 28, available: 24)0You can turn this error into a warning by using the flag ignore_longer_outputs_than_inputs
Following worked for me
Go to
DeepSpeech/training/deepspeech_training/train.py
Now look for following particular line (Normally in 240-250)
total_loss = tfv1.nn.ctc_loss(labels=batch_y, inputs=logits, sequence_length=batch_seq_len)
Change it to as following
total_loss = tfv1.nn.ctc_loss(labels=batch_y, inputs=logits, sequence_length=batch_seq_len, )
Related
I am trying to use zbarcam as a qr code reader.
Build System: Buildroot
Hardware: NXP iMX8MM custom board.
Camera: Omnivision camera
Camera driver was ported successfully and I can run it successfully. I am getting the following error message when I am running the zbarcam app.
Can someone please help me with how to proceed with debugging ?
# zbarcam --verbose=9
_zbar_video_open: opened camera dunknown pixelformat:' '
evice /dev/video0 (fd=5)
_zbar_vmx6s-csi 32e20000.csi1_bridge: Fourcc format (0x00000000) invalid.
unknown pixelformat:' '
4l2_probe: i.MX6S_CSI on platformmx6s-csi 32e20000.csi1_bridge: Fourcc format (0x00000000) invalid.
:32e20000.csi1_bridge driver mx6s-csi (version 5.4.24)
_zbar_v4l2_probe: capabilities: CAPTURE READWRITE STREAMING
v4l2_reset_crop: crop bounds: 0 x 0 # (0, 0)
v4l2_reset_crop: current crop win: 0 x 0 # (0, 0) aspect 1 / 1
v4l2_probe_formats: enumerating supported formats:
v4l2_probe_formats: [0] RGB3 : RGB3 EMULATED
v4l2_probe_formats: [1] BGR3 : BGR3 EMULATED
v4l2_probe_formats: [2] YU12 : YU12 EMULATED
v4l2_probe_formats: [3] YV12 : YV12 EMULATED
v4l2_probe_formats: Max supported size: 0 x 0
v4l2_probe_formats: Found 0 formats and 4 emulated formats.
v4l2_probe_formats: current format: (00000000) 0 x 0 INTERLACED (line=0x0 size=0x0)
v4l2_probe_formats: setting requested size: 40960 x 30720
v4l2_probe_formats: set FAILED...trying to recover original format
v4l2_probe_formats: final format: (00000000) 0 x 0 INTERLACED (line=0x0 size=0x0)
WARNING: zbar video in v4l2_probe_iomode():
system error: USERPTR failed. Falling back to mmap: Invalid argument (22)
_zbar_v4l2_probe: using I/O mode: MMAP
ERROR: zbar processor in _zbar_processor_open():
unsupported request: not compiled with output window support
ERROR: zbar processor in zbar_processor_init():
system error: spawning input thread: Invalid argument (22)
ERROR: zbar processor in zbar_processor_init():
system error: spawning input thread: Invalid argument (22)
Thanks,
Avijit
I am trying to debug my swift program using lldb in terminal. I have compiled my code with the command
swiftc hello.swift
then I ran lldb using
lldb hello
in the lldb program I tried to set a break point using
(lldb) breakpoint set --file hello.swift --line 10
but I got the error
Breakpoint 2: no locations (pending).
WARNING: Unable to resolve breakpoint to any actual locations.
here is the code for hello.swift:
func answer()-> Int{
var value = 12
value = value + 9
return 4 * value
}
let arr = [0, 1, 2, 3, 4]
for _ in arr{
print(answer())
}
what am I doing wrong?
You should set the -g option when compiling the program. From swiftc --help:
-g Emit debug info. This is the preferred setting for debugging with LLDB.
Example session:
% swiftc -g hello.swift
% lldb hello
(lldb) target create "hello"
Current executable set to '/tmp/x/hello' (x86_64).
(lldb) breakpoint set --file hello.swift --line 10
Breakpoint 1: where = hello`main + 256 at hello.swift:10:11, address = 0x0000000100003bf0
(lldb) run
Process 15967 launched: '/tmp/x/hello' (x86_64)
Process 15967 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.1
frame #0: 0x0000000100003bf0 hello`main at hello.swift:10:11
7 let arr = [0, 1, 2, 3, 4]
8
9 for _ in arr{
-> 10 print(answer())
11 }
Target 0: (hello) stopped.
I am currently trying to learn swift and while I was coding some dummy stuff, I noticed the xcode is skipping some lines and sometimes stops on weird lines . my env: macos sierra,xcode Version 8.2.1 (8C1002)
since xcode is skipping some lines , I thought the problem is from code optimisation , then I switched to terminal to debug, below is the input&output
I compiled using
swiftc -g -Onone *.swift
then
load it with lldb
lldb main
set breakpoint on lint 18 as shown below and process run with "r"
17 do {
-> 18 try StaticM.teststatic2()
19 print (1)
20 }catch {
21
(lldb) thread step-in
Process 963 stopped
* thread #1: tid = 0x86e1, 0x0000000100001f6a main`static StaticM.teststatic2(self=main.StaticM, $error=Error # 0x00007fff5fbffac0) throws -> () + 26 at staticExt.swift:13, queue = 'com.apple.main-thread', stop reason = step in
frame #0: 0x0000000100001f6a main`static StaticM.teststatic2(self=main.StaticM, $error=Error # 0x00007fff5fbffac0) throws -> () + 26 at staticExt.swift:13
10 public extension StaticM {
11 #discardableResult
12 public static func teststatic2() throws {
-> 13 var asdf=2;
14 let sdfgsdfg=2;
15 print(sdfgsdfg);
16 print (asdf);
(lldb) n
Process 963 stopped
* thread #1: tid = 0x86e1, 0x0000000100001f72 main`static StaticM.teststatic2(self=main.StaticM, $error=Error # 0x00007fff5fbffac0) throws -> () + 34 at staticExt.swift:15, queue = 'com.apple.main-thread', stop reason = step over
frame #0: 0x0000000100001f72 main`static StaticM.teststatic2(self=main.StaticM, $error=Error # 0x00007fff5fbffac0) throws -> () + 34 at staticExt.swift:15
12 public static func teststatic2() throws {
13 var asdf=2;
14 let sdfgsdfg=2;
-> 15 print(sdfgsdfg);
16 print (asdf);
17 asdf += 1;
18 }
(lldb) process continue
Process 963 resuming
2
2
1
Process 963 exited with status = 0 (0x00000000)
(lldb)
question is , after I go to next line using "n" in lldb, why lldb skips line 14 of staticExt.swift and jumps straight to line 15 ?
also , sometimes when trying to debug some other program, I click on "step-in" in xcode, it stops on the declaration line of the func instead of first line in the code block, I clicked stepover , it goes back to the caller func line instead of executing the first line of that func.
in all,programe works , but why lldb jumps even with -Onone and -g ?
could you please let me where can I find more info ?
thank you very much.
Swift's basic types (Int's, Strings, etc.) are formally somewhat heavy-weight - for instance in your example:
(lldb) expr --raw -- asdf
(Swift.Int) $R0 = {
_value = 3
}
To make performance acceptable even though this is true, the swift compiler "unboxes" these types in common operations and applies other "optimization-like" tricks to make all this go faster. Since this is such a pervasive feature of swift, it does so even at -Onone, to make performance of unoptimized code acceptable. And once the optimizer starts getting involved, it sometimes can't help itself and will do more work than maybe it should at -Onone.
In this case, because the first variable is a "let" of an Int, swiftc knows it can just directly insert the value into the arguments when it goes to call print, so it doesn't need to make up a variable. If you change the let to a var then code will actually be generated for that line, and the line will get its own line table entry.
If you know how to read assembly code, you can look at the mixed disassembly to see this in action.
In lldb:
(lldb) dis -m -f
will give you mixed disassembly of the current frame.
|................................................................ | 98%
|.................................................................| 100%
output file: test.knit.md
/usr/lib/rstudio-server/bin/pandoc/pandoc +RTS -K512m -RTS test.utf8.md --to html --from markdown+autolink_bare_uris+ascii_identifiers+tex_math_single_backslash --output test.html --smart --email-obfuscation none --self-contained --standalone --section-divs --table-of-contents --toc-depth 3 --template /usr/lib64/R/library/rmarkdown/rmd/h/default.html --variable 'theme:cerulean' --include-in-header /tmp/Rtmp4CjD4R/rmarkdown-str6f6421e357bc.html --mathjax --variable 'mathjax-url:https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML' --highlight-style haddock
Error: pandoc document conversion failed with error 127
Occasionally, I run the codes and it completed 100% and created a md format file.
> traceback()
10: eval(expr, envir, enclos)
9: FUN(X[[i]], ...)
8: lapply(aesthetics, eval, envir = data, enclos = plot$plot_env)
7: f(..., self = self)
6: l$compute_aesthetics(d, plot)
5: f(l = layers[[i]], d = data[[i]])
4: by_layer(function(l, d) l$compute_aesthetics(d, plot))
3: ggplot_build(x)
2: print.ggplot(x)
1: function (x, ...)
UseMethod("print")(x)
There is no ggplot function run inside the RMarkdown but there is ggplot error when I tried to traceback(). However there is an error Error:127, any solution?
Googling the error message indicates that ‘status 127’ means that commands are not being found.
I would like to read .yml files in Matlab. These files contain coordinates x and y of key points on a face image. I looked for different tools but I don't seem to find any answers.
My .yml files look like this
YAML:1.0
Image file: "00032009.jpg"
Contours count: 8
Contours:
-
Name: FO
Count: 41
Closed: 0
Points:
-
x: 682.5947265625000000
y: 743.1998901367187500
-
x: 685.9638061523437500
y: 771.3800659179687500
......
and so on
Note 00032009.jpg is an image of a face
x and y are coordinates of a point on a face Eg: the right corner of an eye etc
Could you please point out a way to read the file and then display the points on the face image?
Edit 1: Here is the error I get
Error: File: ReadYamlRaw.m Line: 14 Column: 11
Expression or statement is incorrect--possibly unbalanced (, {, or [.
Error in ==> ReadYaml at 38
ry = ReadYamlRaw(filename, 0, nosuchfileaction);
What is weird is line 14 in ReadYamlRaw.m
[pth,~,~]= fileparts(mfilename('fullpath'));
Parse error at ','(second one) and ']' usage appears to be invalid matlab syntax.
So what is the use of ~ in there and why is there an error?
Edit2: I replaced the ~ in the line above with dummy variables then I get this errors O_O
Error using ==> ReadYamlRaw>scan at 81
Unknown data type: logical
Error in ==> ReadYamlRaw>scan_map at 138
result.(ich) = scan(r.get(java.lang.String(ich)));
Error in ==> ReadYamlRaw>scan at 79
result = scan_map(r);
Error in ==> ReadYamlRaw>scan_list at 116
result{ii} = scan(i);
Error in ==> ReadYamlRaw>scan at 77
result = scan_list(r);
Error in ==> ReadYamlRaw>scan_map at 138
result.(ich) = scan(r.get(java.lang.String(ich)));
Error in ==> ReadYamlRaw>scan at 79
result = scan_map(r);
Error in ==> ReadYamlRaw>load_yaml at 48
result = scan(yaml.load(fileread([filename, fileext])));
Error in ==> ReadYamlRaw at 19
result = load_yaml(filename, nosuchfileaction);
Error in ==> ReadYaml at 38
ry = ReadYamlRaw(filename, 0, nosuchfileaction);
I also tried with a different yml file that looks like this
%YAML:1.0
RE-C:
x: 919
y: 580
LE-C:
x: 1209
y: 597
N-C:
x: 1063
y: 698
FO-B:
x: 1045
y: 1114
REL-O:
x: 852
y: 597
REL-I:
x: 986
y: 600
REL-T:
x: 918
y: 564
And I get the following errors
Java exception occurred:
while scanning a directive
in "<string>", line 1, column 1:
%YAML:1.0
^
expected alphabetic or numeric character, but found :(58)
in "<string>", line 1, column 6:
%YAML:1.0
^
at org.yaml.snakeyaml.scanner.ScannerImpl.scanDirectiveName(ScannerImpl.java:1028)
at org.yaml.snakeyaml.scanner.ScannerImpl.scanDirective(ScannerImpl.java:990)
at org.yaml.snakeyaml.scanner.ScannerImpl.fetchDirective(ScannerImpl.java:534)
at org.yaml.snakeyaml.scanner.ScannerImpl.fetchMoreTokens(ScannerImpl.java:251)
at org.yaml.snakeyaml.scanner.ScannerImpl.checkToken(ScannerImpl.java:179)
at
org.yaml.snakeyaml.parser.ParserImpl$ParseImplicitDocumentStart.produce(ParserImpl.java:198)
at org.yaml.snakeyaml.parser.ParserImpl.peekEvent(ParserImpl.java:161)
at org.yaml.snakeyaml.parser.ParserImpl.checkEvent(ParserImpl.java:146)
at org.yaml.snakeyaml.composer.Composer.getSingleNode(Composer.java:105)
at org.yaml.snakeyaml.constructor.BaseConstructor.getSingleData(BaseConstructor.java:121)
at org.yaml.snakeyaml.Yaml.loadFromReader(Yaml.java:480)
at org.yaml.snakeyaml.Yaml.load(Yaml.java:399)
Error in ==> ReadYamlRaw>load_yaml at 48
result = scan(yaml.load(fileread([filename, fileext])));
Error in ==> ReadYamlRaw at 19
result = load_yaml(filename, nosuchfileaction);
Error in ==> ReadYaml at 38
ry = ReadYamlRaw(filename, 0, nosuchfileaction);
Maybe someone can make something out of these or you could point out another set of functions that would work? I searched but didn't find any except this one.
It's YAML file indeed (as #DavidBrown mentioned in his comment, the extension does not matter). But it has some problems. Don't know if it's due to wrong YAML format or MATLAB implementation.
I've installed YAMLMATLAB and played a little with your file.
YamlStruct = ReadYaml(yaml_file);
YAMLMATLAB returns error if the files is feed as is. It works only if I comment the first line and remove spaces from field names. So the beginning of the file looks like this:
#YAML:1.0
Imagefile: 00032009.jpg
Contourscount: 8
...skipped the rest...
Then I get the correct structure. For example you can access the 1st point's x coordinate as
YamlStruct.Contours{1}.Points{1}.x
ans =
682.5947
UPDATE
Space in filed names is actually a known problem in YAMLMATLAB. See the report and possible solution here.
UPDATE 2
According to comment from #Jirka_cigler (YAMLMATLAB developers group):
In the program release 0.4.3 we added support for whitespaces in field
names, so the problem should not appear again.
Great!
I've also removed previous developers comment on the problem in the first UPDATE since it's not true any more.
ZozaniDB Database Toolbox comes with a Matlab-native YAML implementation that handles the .yml file here. To parse a string or a file, use:
>> yaml_parse ( str )
>> yaml_read ( filename )