quickTunerPreproc.py, preprocessor script for quick tuner scripts #1575
Conversation
This script accepts the *.debug files that are generated from tuningRunner.py as input and concatenates them into a single tsv file for use in the next step of the quick tuning pipeline: generation.
…pts." This reverts commit 0489788.
| def __init__(self, pargs): | ||
| self.input_dir = pargs.input_dir | ||
|
|
||
| @staticmethod |
There was a problem hiding this comment.
Why staticmethod? Not a criticism, I'm just less familiar with why that's done in python and I'm curious.
There was a problem hiding this comment.
This was intended to be imported and called within the quickTunerGen.py set of classes but I could not rationalize each individual class having their own instance of this preprocessor class hanging out inside:
So, it was either have a standalone function or a class with a static method that would allow me to do something like:
qtPreprocessor.process( /* args */ )
From my experience with using the @staticmethod decorator, it has always been what to use when you want to package functions together that have similar functionality and can share data but do not share state.
| df = pd.read_csv(file, sep='\t') | ||
| if normalize: | ||
| scaler = MinMaxScaler() | ||
| df['TFlops'] = scaler.fit_transform(df[['TFlops']]) |
There was a problem hiding this comment.
Why one set of brackets on the left but two on the right??
There was a problem hiding this comment.
I think that was a mess up on my end. Fortunately (or unfortunately) the syntax is forgiving.
Also, this will actually be changed to add a NormalizedTFlops column instead of overwriting the current one.
if normalize:
scaler = MinMaxScaler()
df['NormalizedTFlops'] = scaler.fit_transform(df[['TFlops']])
|
|
||
| to: | ||
|
|
||
| import uuid |
There was a problem hiding this comment.
Could this instead be a patch to rocmlir_worker?
There was a problem hiding this comment.
Not in its current form; it cuts off the usual functionality of passing results to stdout and we don't have code to read results from the files and send it on. I do have an open issue about collecting all the results into the database and this may become part of it.
| import glob | ||
| from sklearn.preprocessing import MinMaxScaler | ||
|
|
||
| class qtPreprocessor(object): |
There was a problem hiding this comment.
Nit class case, and I'm pretty sure Python 3 doesn't require the (object) anymore
|
|
||
| dfs = [] | ||
| ct = 0 | ||
| for file in tsv_files: |
There was a problem hiding this comment.
You can emumerate() instead of maintaining a ct
There was a problem hiding this comment.
I will replace this with a len() call since it is not needed within the loop only for generating stats.
counter for processing loop.
| import glob | ||
| from sklearn.preprocessing import MinMaxScaler | ||
|
|
||
| class qtPreprocessor(): |
There was a problem hiding this comment.
Standard complaining re style on class names
djramic
force-pushed
the
qtPreprocessor
branch
from
1eac2a7
to
30999f4
Compare
This script accepts the *.debug files that are generated from tuningRunner.py as input and concatenates them into a single tsv file for use in the next step of the quick tuning pipeline: generation.