You guys know how to bebug already. We all do. But when it comes to Machine Learning and Tensorflow or the likes, things are getting annoying if we were still using traditional debuggers. A programmable debugger is what in my mind and probably in yours too. One breakpoint to investigate about everything! At this point, you can then test whatever you want, supported by all the power of FORTH.
So now we need to choose an interactive UI and its syntax that is light weight, reliable and flexible so we won't regret of choosing it someday, has been there for decades so many people don't need to learn about another new language although we are only to use some debug commands, yet easy enough for new users, that's FORTH.
An easy way to try peforth is through the Microsoft Azure Notebooks. Create a Jupyter notebook page (try mine) and install peforth on that page with this line:
!pip install peforth
# If you want to install a specific version
!pip install peforth==1.23
# "Clear before install" is a good idea. Give it a try if things went wrong.
!pip uninstall -y peforth
and then you can import peforth like this:
import peforth
and then start using peforth through magics %f and %%f as shown below:
You can even enter the peforth interprete state to 'talk' with it:
%f *debug* foo>
don't foget to use 'exit' command to stop talking or the page will concentrate on the dialog and not to do anything else.
pip install peforth
import peforth on Jupyter Notebook is the only thing you need to do to use peforth
%f and %%f magics. For tutorials, please find and read jupyter notebooks in the 'notebook' directory of this project.
Print "Hello World!"
Microsoft Windows [Version 10.0.15063]
(c) 2017 Microsoft Corporation. All rights reserved.
c:\Users\your-working-folder>python -m peforth .' Hello World!!' cr bye
Hello World!!
c:\Users\your-working-folder>
so your peforth is working fine.
To your application, import peforth as usual to bring in the debugger:
c:\Users\your-working-folder>python
Python 3.6.0 (v3.6.0:41df79263a11, Dec 23 2016, 08:06:12) [MSC v.1900 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import peforth
p e f o r t h v1.07
source code http://github.com/hcchengithub/peforth
Type 'peforth.ok()' to enter forth interpreter, 'exit' to come back.
>>>
The greeing message tells us how to enter the FORTH interpreter for your debugging or investigating and how to come back to continue running your code.
# 100.py
sum = 0
for i in range(100):
sum += i
print("The sum of 1..100 is ", sum)
Run it:
c:\Users\your-working-folder>python 100.py
The sum of 1..100 is 4950
c:\Users\your-working-folder>
The result should be 5050 but it's not! Let's drop a breakpoint to see what's wrong:
# 100.py with breakpoing .----- Specify an unique command prompt to indicate where
| the breakpoint is from if there are many of them
import peforth | .----- pass locals() at the breakpoint
sum = 0 | | to our debugger
for i in range(100): | | .------- use a FORTH constant
sum += i | | | to represent the locals()
peforth.ok('my first breakpoint> ',loc=locals(),cmd="constant locals-after-the-for-loop")
print("The sum of 1..100 is ", sum)
Run again:
c:\Users\your-working-folder>python 100.py
p e f o r t h v1.07
source code http://github.com/hcchengithub/peforth
Type 'peforth.ok()' to enter forth interpreter, 'exit' to come back.
.--------------- at the breakpoint, type in 'words'
| command to see what have we got
my first breakpoint> words .-------- It's a long list of 'words'
... snip ....... | or available commands. Don't worry, we'll use only some of them.
expected_rstack expected_stack test-result [all-pass] *** all-pass [r r] [d d] [p
p] WshShell inport OK dir keys --- locals-after-the-for-loop
|
The last one is what ------'
we have just created throuth the breakpoint statement
, named "locals-after-the-for-loop"
Let's see it:
print a carriage return at the end -------.
print the thing -----. |
| |
my first breakpoint> locals-after-the-for-loop . cr
({'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__':
<_frozen_importlib_external.SourceFileLoader object at 0x000001DD2D737710>,
'__spec__': None, '__annotations__': {}, '__builtins__': <module 'builtins'
(built-in)>, '__file__': '100.py', '__cached__': None, 'peforth': <module 'peforth'
from 'C:\\Users\\hcche\\AppData\\Local\\Programs\\Python\\Python36\\lib\\site-packages\\pe
forth\\__init__.py'>, 'sum': 4950, 'i': 99}, {}, 'my first breakpoint> ')
my first breakpoint> | | |
| | '--- our command
our sum -----' | prompt
| indicates where the
99 instead of 100 ----------' breakpoint is from
this is the problem !!
Now leave the breakpoint and let the program continue:
my first breakpoint> exit
my first breakpoint> The sum of 1..100 is 4950
c:\Users\your-working-folder>
When at a breakpoint in Tensorfow tutorials, I always want to make some experiments on those frustrating tf.something(tf.something(...),...) things to have a clearer understanding of them without leaving the underlying tutorial. Let's use the above example again in another way to demonstrate how to do that with peforth:
Run peforth:
Microsoft Windows [Version 10.0.15063]
(c) 2017 Microsoft Corporation. All rights reserved.
c:\Users\hcche\Downloads>python
Python 3.6.0 (v3.6.0:41df79263a11, Dec 23 2016, 08:06:12) [MSC v.1900 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import peforth
p e f o r t h v1.07
source code http://github.com/hcchengithub/peforth
Type 'peforth.ok()' to enter forth interpreter, 'exit' to come back.
>>> peforth.ok()
OK <-------- Default FORTH command prompt
OK
Assume we are at a breakpoint and we need a procedure to
add 1..100 to get the sum of them. We are not sure if the procedure
is correct so we need to try. Now copy the procedure from
your text editor. The <py>...</py> tells the debugger that
the code within is a block of in-line python.
The outport() function outports the given locals() to the
FORTH environment outside the in-line python block.
<py>
sum = 0
for i in range(100):
sum += i
print("The sum of 1..100 is ", sum)
outport(locals())
</py>
It's a block of multiple-line text strings so we press Ctrl-D to start a multiple-line input, copy-paste, and press another Ctrl-D to end the multiple-line block. Like this:
OK
OK ^D
<py>
sum = 0
for i in range(100):
sum += i
print("The sum of 1..100 is ", sum)
outport(locals())
</py>
^D
The sum of 1..100 is 4950
OK
Now use the 'words' command to see what have we got:
OK words
code end-code \ // <selftest> </selftest> bye /// immediate stop compyle
trim indent -indent <py> </py> </pyV> words . cr help interpret-only
compile-only literal reveal privacy (create) : ; ( BL CR word ' ,
[compile] py: py> py:~ py>~ 0branch here! here swap ! @ ? >r r> r@ drop
dup over 0< + * - / 1+ 2+ 1- 2- compile if then compiling char last
version execute cls private nonprivate (space) exit ret rescan-word-hash
(') branch bool and or not (forget) AND OR NOT XOR true false "" [] {}
none >> << 0= 0> 0<> 0<= 0>= = == > < != >= <= abs max min doVar doNext
depth pick roll space [ ] colon-word create (marker) marker next abort
alias <> public nip rot -rot 2drop 2dup invert negate within ['] allot
for begin until again ahead never repeat aft else while ?stop ?dup
variable +! chars spaces .( ." .' s" s' s` does> count accept accept2
<accept> nop </accept> refill [else] [if] [then] (::) (:>) :: :> ::~
:>~ "msg"abort abort" "msg"?abort ?abort" '<text> (<text>) <text> </text>
<comment> </comment> (constant) constant value to tib. >t t@ t> [begin]
[again] [until] [for] [next] modules int float drops dropall char>ASCII
ASCII>char ASCII .s (*debug*) *debug* readTextFile writeTextFile
tib.insert sinclude include type obj>keys obj2dict stringify toString
.literal .function (dump) dump dump2ret d (see) .members .source see dos
cd slice description expected_rstack expected_stack test-result
[all-pass] *** all-pass [r r] [d d] [p p] WshShell inport OK dir keys
--- i sum
OK
At the end of the long list after the --- marker we found i and
sum. They are all locals() at the point in the in-line python block.
Let's see them:
OK i . cr
99
OK sum . cr
4950
OK
Again, we found the root cause of why the sum is not 5050 because
i didn't reach to 100 as anticipated. That's exactly how the
python range() works and that has actually confused me many times.
Visit this project's Wiki pages for more examples about how to view MNIST handwritten digit images at the half way of your investigating in a Tensorflow tutorial, for example, and the usages of this programmable debugger.
H.C. Chen, FigTaiwan
[email protected]
Just undo it!