Outreachy/MDF task

[Olawumi-11]

This is the branch I created for my MDF task. In this branch, I created a repository called Outreachy_Modeci_Projects where I created a separate folder for each of my task. In each folder is a Readme.md file that described the outline of each of this task. I humbly await your feedback on all this tasks. Thanks

[pgleeson]

Thanks for this @TaofeeqatDev.

• For the pytorch to mdf example, please make sure you have run it with the latetst version of mdf, which gives colors to the generated graph of the mdf model. Aslo, use the new solid_color=True option in mdf_model.to_graph_image
• Can you run the model multiple times with different input digits to assess the % accuracy of the MDF model?
• The keras_mdf.py file looks like it comes from a previous commit, which is not in our main branch. Can you be clear where this came from?
• Can you add a new layer to your keras model and retrain it, and demonstrate that the newly generated mdf model also has the new structure? Note you should use these helper methods: https://github.com/ModECI/MDF/blob/main/examples/TensorFlow/Keras/keras2mdf.py

[Olawumi-11]

@pgleeson As regards the last feedback on my project, I have added a new layer to my keras model as instructed and have also retrained it and generated the graph. I'm currently working on the helper functions you referred to in order to convert it to MDF.

[pgleeson]

Thanks @TaofeeqatDev.

• It seems though some of your notebooks have disappeared in this version, particularly where the pytorch and keras generated models are converted to mdf.
• Can you remove the files used to generate the images with graphviz from this commit (e.g. AB, example - text files which begin with digraph)?
• Can you tell me where this file came from originally?

[Olawumi-11]

Thanks @pgleeson

• I removed the 2 notebooks in order to work on the changes you talked about base on your last feedback.
• I will remove the files from the folder
• I just confirmed the source of the script, it's from an ex-intern repository. I came across it while reviewing previous work that had been done. Since you already referred me to a new script, I'm currently working with that. Thanks

[Olawumi-11]

@pgleeson as regards your comment on too many files in PR. My question is, do you want to upload just the two notebook, I mean the notebook for building the model and the one for converting to MDF. Just want to confirm this before I reupload my projects.
Do we also need to include the link to our pull request when recording our contribution on outreachy website or just the link to the project folders is enough?

[Olawumi-11]

@pgleeson as regards the last feedback on my task.

• I have added a new layer to my keras model and retrained it.
• I also use the helper functions you referred to for my keras to MDF conversion.
• I'm currently working on the pytorch to MDF conversion and I will reupload it soon.

While working on the pytorch to MDF conversion I supply different input data to test the accuracy of my model. Sometimes the keras model output will be exactly the same as MDF, and sometimes there is usually a slight difference between their output.

[pgleeson]

@TaofeeqatDev there shouldn't be any reason to remove files and reupload them, e.g. your notebooks. You should be managing files by adding/committing changes using git, rather adding/deleting them on the website.
Please do add the pytorch->mdf version and update that by committing changes to the same file.
Re the keras example, it seems to be just a toy example with dummy data in/out, correct? It would be better if it was a "real world" example, e.g. like the pytorch one
The file referred to in your keras notebook (from Keras_MDF import *) does not exist.

[Olawumi-11]

@TaofeeqatDev there shouldn't be any reason to remove files and reupload them, e.g. your notebooks. You should be managing files by adding/committing changes using git, rather adding/deleting them on the website. Please do add the pytorch->mdf version and update that by committing changes to the same file. Re the keras example, it seems to be just a toy example with dummy data in/out, correct? It would be better if it was a "real world" example, e.g. like the pytorch one The file referred to in your keras notebook (from Keras_MDF import *) does not exist.

Thanks for the feedback... I will use the MNIST dataset that I use for my pytorch example

[Olawumi-11]

I have opened a new pull request for my project. Thanks