pCDs – polarisCodingDays

ML information can be found here –> JP’s ML Corner

#018 2022-09-08 pytest continues

• pip install tox
• $ sudo apt install git-lfs
• $ git lfs install
• $ git lfs pull

#017 2022-09-07 pytest into it

• diving back into it again
• setting up the dev env was a tricky issue
• Effective Python Testing With Pytest – RealPython
• order how polaris calls functions
  • polaris
  • polaris-fetch
  • data_fetch_decode_normalize
• fixed pytest in tox.ini -> 7.1.3
• pytest error in test_data_normalize_skip_normalizer(datafiles)

#016 2020-05-22 CoronaCoding#015

016-1 improving test input for test.find_best_method()

• getting back in the game

016-2 bugs in automated feature selection

• P1: The number of selected features varies by almost every run between 160 and 327
  I suspect the randomness in train-test-split as the root of this but not completely sure since it is just switching between these two
• P2: The count of selected features in variable does not match the number of variables in mlflow
  Maybe it has something to do with double values … not sure
  • len(feature_list) is not number of selected features in fdf
  • there might be doubles in feature_list
  • best_features_tmp = feature_list
    should be
    best_features_tmp = $features in fdf$
• P3: the output of the method “all_best“` are just 2 features. This is absolutely not the intention

#015 2020-04-06 CoronaCoding#014

015-1 pytest #72

• using pytest.fixture to create “global” variables for more than one pytest
• pytest.fixture() [pytest Doc]
• pandas.DataFrame [pandas doc]
• pandas.DataSeries [pandas doc]

#014 2020-04-04 CoronaCoding#013

014-1 pytest #72

• using VSCode debugger to test find_best_method()
• I believe in order to test find_best_method() it is necessary to test creat_feat_select_df() first DONE
  • problem with defining a DataFrame as the input for the pytests
• using “imaginary” data for testing is quite helpful
• radom.randrange() (Python3 Doc)

#013 2020-04-03 CoronaCoding#012

013-1 Finishing the work on #72

• Parameterizing Tests (Pytest Doc)
• writing a pytest for find_best_method()
  • ASSUMPTION
    being able to use the VSCode properly will help to define pytests
    • both need a defined input for the file/method to test

#012 2020-04-02 CoronaCoding#011

012-1 Finishing the work on #72

• writing a pytest

#011 2020-03-31 CoronaCoding#010

011-1 Finishing the work on #72

• I could resolve all problems and issues
• I re-ordered the code
• I excluded the creation of a DataFrame solely filled with selected features into its own function create_feat_select_df()
• problem with the pytest for XCorr regarding the creation of an experiment
  • implemented the experiment creation process into the pytest itself

#010 2020-03-30 CoronaCoding#009

010-1 Continuing on #72

• these notes stand as an example for the last days
• activating the feature selection process
  • used the reader method to use the combined json as an input for the feature extraction
• polaris fetch -s 2019-08-10 -e 2019-09-5 LightSail-2 /tmp/normalized_frames.json
• polaris learn -g /tmp/new_graph.json /tmp/normalized_frames.json -c bat0_temp
• The logging and selection method rotation seem not to work properly since the feature seems to have an empty field
  • “mlflow.exceptions.MlflowException: Got invalid value [] for metric ‘Feature List’ (timestamp=1585565094701). Please specify value as a valid double (64-bit floating point)“
• The model needs to be trained with the selected features
  • TODO Create DataFrame with the selected features <–DONE
  • Has something like the been done?
• problem with duplicates in first_best <– LATER

#009 2020-03-36 CoronaCoding#008

009-1 Continuing on #72

• trying to figure out why my implementation seems not to be evoked
  • the feature selection process was not being called

#008 2020-03-26 CoronaCoding#007

008-1 Continuing on #72

• Had some problems regarding the debugger in VSCode. The break point was not visible.
  I ended up finding the editor.glyphMargin in the settings.json and switched it from false to true
  • This one helped – stackoverflow
  • Python debug configurations in Visual Studio Code – VSCode
• I resolved all tox errors but one
• It seems that the implemented functionality is not being executed at the moment. The search for the cause is still ongoing.

#007 2020-03-23 CoronaCoding#006

007-01 Preparing Slides for Space: Episode II

007-2 Continuing on #72

• cleaning up the current status with tox
• still on the struggle to find a way to compare the logged prediction results
  • Track machine learning training runs – databrick

#006 2020-03-22 CoronaCoding#005

006-01 Preparing Slides for Frankfurt Data Science: Space – Episode II

• PostIdea: Why do be need automated spacecraft diagnostics?

#005 2020-03-21 CoronaCoding#004

005-1 Continuing on #72

• Done train test split to have data for training and testing
• running mlflow ui inside the polaris main directory

005-2 Updating README.md

• ran into problems with the fetch command
• Output is now a file, not a directory
• normalized_frames.json seems to be obsolete

#004 2020-03-20 CoronaCoding#003

004-1 Continuing on #72

• used input_dataset and input_y from selection.fit() as train and test data set for the prediction in mlflow
• problem with venv
  • reboot and updating python to 3.8.2

#003 2020-03-19 CoronaCoding#002

003-1 Continuing on #72

• python – staticmethod
• MLflow
  • ML Experiment Tracking using MLflow (YouTube)
  • MLflow Tutorial (MLflow Doc)
• MLflow – Tracking – experiments : “feat_select_meth_validatoin”
  • multiple runs with a set (prediction&evaluation) for each selection method
• sklearn.metrics.mean_squared_log_error (Doc)

#002 2020-03-18 CoronaCoding#001

• getting back in the game & figuring out what key aspects for #72 are
  • finding a way to perform auto feature selection
    –> new method: selection.py:find_best_method()
• selection.py:find_best_method()
  • 1. predict $something$ with each selection method
  • 2. safe score with mlflow – Tracking
    Choosing..(medium),
  • 3. compare results (RMSE?)
  • 4. pick best selection method
• Where to call find_best_method() ?
  • ? extract_best_features() ? <– Yes
• MLflow-Tracking
  • run: “executes data science code”
  • experiment: organizes multiple runs

#001 2020-02-08 Saturday Coding Adventure @ TUDSaT

• learn: FeatureSelectionProgress
  • figuring out:
    • where to select the methods
      Using list for pipelining each element?
      –> This could be editable on a higher level.
    • Can FeatureImportanceOptimization be used for finding the best feature selection method?
• viz: node colours
  • created a colour palette
  • List of 20 Simple, Distinct Colors