dxc-technology / dxc-industrialized-ai-starter Goto Github PK

Describe the area of code that needs more transparency:
Display metrics or statistics that show the difference between raw data and clean data.
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
Show the stats of raw and clean data as different columns
We should have metrics for categorical and numerical data, should also think about how to handle providing usable metrics for data sets with lots of features.

Describe the area of code that needs more transparency:
Provide Auto-ML documentation link in the user guide for running models
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
If possible we should provide an easy way to expose deep links to the specific algorithms as part of our way to support Data Scientist making their work explainable

Describe the bug
distplot is a deprecated function and will be removed in a future version, we have to find an alternative to replace this function

To Reproduce
Steps to reproduce the behavior:

1. in the DXC-Industrialized-AI-Starter.ipynb in colab, execute this command ai.plot_distributions(data1)
2. you will get this warning
   distplot is a deprecated function and will be removed in a future version. Please adapt your code to use either displot (a figure-level function with similar flexibility) or histplot (an axes-level function for histograms). Pass the following variable as a keyword arg: x. From version 0.12, the only valid positional argument will be data, and passing other arguments without an explicit keyword will result in an error or misinterpretation.

Expected behavior
WE have to find an alternative for this function

Screenshots

Additional context
N/A

I am just wonder how do i test the api keys obtained from Mongodb using this library. Is it included in it ?

Where i refer to test what it meant is to check if the api keys are working and able to successfully establish a connection.

Describe the area of code that needs more transparency:
Add encryption to the published microservice
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
research levels of best-practice security for different types of data. We could offer a parameter mapped to predefined configurations for low, medium, high, and/or extra-high levels of security.

Describe the bug
Trying to upload an excel file (in Colab) but its failing with below error

read_data_frame_from_local_excel_file()
29 uploaded = files.upload()
30 excel_file_name = list(uploaded.keys())[0]
---> 31 df = pd.read_excel(io.BytesIO(uploaded[excel_file_name]))
32 return(df)
33

NameError: name 'io' is not defined

To Reproduce
Steps to reproduce the behavior:

1. dataframe = ai.read_data_frame_from_local_excel_file()
2. Browse the excel file (in Colab)
3. See error

Expected behavior
Dataframe with the Excel data

Screenshots
If applicable, add screenshots to help explain your problem.

Additional context
Add any other context about the problem here.

Publish DXC-Industrialized-AI-Starter google colab notebook file to GitHub.

By default ai.run_experiment() produces a lot of output. Can we add a parameter to the function that defines a verbose and succinct mode? In verbose mode, the function produces the full output, but in succinct mode, only a summary is output. I recommend making succinct mode the default.

Users in the boot camp is facing "as_matrix" error when calling ‘ai.explore_features(raw_data)’ in the AI Workbook.

The module is deprecated in version 0.21 and removed in version 0.23. So hotfix need to be done to avoid problem in boot camp usage.

Problem:
In the DXC_Industrialized_AI_Starter.ipynb notebook, there are a couple instances where it suggests you can upload a local file to colab. This doesn't work. The error message I get is :

MessageError: TypeError: Cannot read property '_uploadFiles' of undefined

MM/D/YYYY and MM/DD/YYYY are two popular US date format and it appears the ai.clean_dataframe() function cannot recognize them.

ParserError: Could not match input '10/1/2020' to any of the following formats: YYYY-MM-DD, YYYY-M-DD, YYYY-M-D, YYYY/MM/DD, YYYY/M/DD, YYYY/M/D, YYYY.MM.DD, YYYY.M.DD, YYYY.M.D, YYYYMMDD, YYYY-DDDD, YYYYDDDD, YYYY-MM, YYYY/MM, YYYY.MM, YYYY, W

Describe the area of code that needs more transparency:
Add test cases links in contributing guide, revisit the contribute guide document, and do the necessary changes.
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:

Most of those attending the DXC-Industralized AI course have not used Colab before.
Some of the instructions for getting started in the "Set up the development environment" could be clearer.
e.g. "This code installs all the packages you'll need. Run it first."
Most people who have not used Colab before would not know how to do this.

can you include a time series model option in the experiment design?

Describe the area of code that needs more transparency:
indicate the completeness or correctness of the data and show the outliers
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
Logan comments:
Not just visualization, completeness, correctness, outliers, and other metrics should be saved statistics, too. We can’t force others to make their work explainable to an end-user, but we can ensure that their work is capable of being explained if they use our package.

Refactor the initial version of the code into sub modules.

Generalizing the conversion of Column names issue in clean_dataframe, write_raw_data, pipeline functions.

In the documentation can you show an example of publishing a microservice from a custom model instead of a model generated from run_experiment() function? The custom model could be something as simple as a function that adds two number or appends text to an input.

User is facing 'int' object has no attribute 'split' error in data pipeline function while working with numeric data operators.

Create an Issue template to capture the issues/changes for AI Ethics & Trustworthiness

The link in the example that is supposed to list available data sets is broken

Describe the bug
Error occurred when trying to import ai from dxc

To Reproduce
Steps to reproduce the behavior:

1. used pip to install ai starter package
2. from dxc import ai
3. error

Expected behavior
ai was getting loaded earlier

Screenshots

When the function is writing source code to algoritima, then there is an issue in that source code.

Describe the bug
I am unable to run AI experiments for Time-Series problems

To Reproduce
When I passed timeseries as a parameter, I got an error as below:

I tried to pick up an example from https://nbviewer.jupyter.org/github/dxc-technology/DXC-Industrialized-AI-Starter/blob/c58754247060262ac0949396e48f71861cb79d4e/Examples/Time_series_Model.ipynb

on setting the value : "model": 'timeseries',
The timeseries values are not displaying as expected. Instead it shows the same value for all predictions

Please let me a way to handle timeseries problems

Expected behavior
Please create a function for Time-Series problem. Please revoke the functionality. As it looks like, it was already implemented

Screenshots
Added the images

Additional context
Add any other context about the problem here.

Resize the pictures
Convert all images into the same file format
Merging the images into a single file
Convert images into a CSV file
Few Changes to the CSV file
Loading the CSV file

ai.read_data_frame_from_local_csv() needs to be able to accept files with other types of separators. I tried to import a pipe delimited csv file and could not set sep = '|' to be able to read the file

Describe the bug
The code fails when I run the part responsible for importing ai from DXC (from dxc import ai)

To Reproduce
Steps to reproduce the behavior:

1. pip install DXC library (First block of code)
2. run the second block of code (from dxc import ai)

Expected behavior
The code should import the ai from DXC library without any issue

Screenshots

Additional context
Non

User facing issue with column names in "access_data_from_pipeline" function in below scenarios:

1. When column names are case sensitive.
2. When column have SPACE in between.

So fix need to be done to:

1. Handle column names case sensitive.
2. Handle column with SPACE in between to replace with “_”.

Describe the bug
the lib pandas.io.json.json_normalize is depreceated, recommendation to use pandas.json_normalize instead

To Reproduce
Steps to reproduce the behavior:

1. Go to build data pipe line in Collab
2. Click on the cell #TODO: Define the code needed to refine the raw data
3. Run cell
4. See error

Expected behavior
no error from Pandas lib

Screenshots

Describe the bug
Pip install DXC_Industrialized_AI_Starter-2.3.9-py3-none-any.whl takes several minutes and times out in Google Colab due to multiple versions of libraries being downloaded

To Reproduce
Steps to reproduce the behavior:
run pip install

Expected behavior
The starter to load in at the most 10 minutes

Screenshots
If applicable, add screenshots to help explain your problem.

Additional context
Add any other context about the problem here.

Can you include a clustering model option in the experiment design?

Reasons to implement

1. User has the flexibility to download only required datasets.
2. Reduce the AI starter library size by moving all datasets to DXC public GitHub and providing user a flexibility to download.

Can you include a deep learning model option in the experiment design?

Describe the area of code that needs more transparency:
Rename the variable name representing the dataset after cleaning the data (ex: Raw_data, Clean_data)
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:

After getting a "rework" status on a badge, doing the changes required and then resubmits gets you an error response that an assertion already exists:

<Response [400]> { "errorMessage": { "statusCode": 400, "exception": "BadRequest", "message": "Assertion already exists", "payload": { "evidence": "https://colab.research.google.com/drive/1oMUEVLS5x1netqWaL0kAt8FVR-ABeJ4U", "lastUpdated": "2021-02-10T15:37:33Z", "badge": "Create a Data Story", "status": "rework", "created": "2021-02-09T16:06:25Z", "comments": [ { "date": "2021-02-10T15:37:33Z", "comment": "Hello, nice work, but for the sample data set area, please upload your data set, not the iris file.", "email": "[email protected]" } ], "email": "[email protected]", "d1": "user:6366a530-391b-41be-bbff-6f372658afef", "d2": "badge:dd05bbdf-ad5b-469d-ab2c-4dd218fd68fe", "salt": "ecaa9028a8feb321be864bf98ac1ebe6", "reviewer": "[email protected]", "sk": "assertion", "pk": "assertion:93372602-81eb-47dc-bf05-5bc475c276b6" } } }

Describe the area of code that needs more transparency:
Add the drift function after pipeline
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
Calculate the drift between given two data sets over a period of time or between training sets

User facing issue with ai.clean_dataframe function when

1. column names are in upper case
2. Having SPACE in between names.

In publish microservice, we need to create a new algorithm if that is not existing in algorithmia. It was not working in our previous code and did changes to make it run.

By default ai.publish_microservice() produces a lot of output. Can we add a parameter to the function that defines a verbose and succinct mode? In verbose mode, the function produces the full output, but in succinct mode, only a summary is output. I recommend making succinct mode the default.

Describe the area of code that needs more transparency:
Include metadata details while writing data into MongoDB
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
Include in the code as a minimum, but best in Mongo
Where did we got the data from, did we ran cleaner on the data
What version of data is
Where we stored the data
This is a manual entry

Describe the area of code that needs more transparency:
Add logs to the crucial steps that give feasibility to the user to revert changes
Describe the solution you'd like:
Describe the alternatives you've considered:
Additional context or comments:
Add logs to the crucial steps(aggregation) which give the feasibility to the user to revert changes if something goes wrong.
Log storage options, recommend config options (at least) for both Mongo Atlas DB and local storage

Add function to read Json file from local system using library and handle cases for nested dictionary Json file.

Do changes so AI_Starter can Integrate with Mongo dB Version 4.2.

Explore LIME, SHAP Libraries and implement Interpretable Machine Learning Models in AI Library.