# Mutaz\_ahmed - Machine Learning Pathway

**URL:** <https://stemaway.com/t/mutaz-ahmed-machine-learning-pathway/4049>\
**Category:** Machine Learning Pipeline (NLP) Self-Assessments\
**Created:** [June 25, 2020, 2:42am UTC](https://stemaway.com/t/mutaz-ahmed-machine-learning-pathway/4049 "2020-06-25T02:42:21Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![mutaz\_ahmed](https://d2nat51eddc6b4.cloudfront.net/user_avatar/stemaway.com/mutaz_ahmed/32/1688_2.png) [@mutaz\_ahmed](https://stemaway.com/u/mutaz_ahmed)\
**Post date:** [June 25, 2020, 2:42am UTC](https://stemaway.com/t/mutaz-ahmed-machine-learning-pathway/4049/1 "2020-06-25T02:42:21Z")

</div>

Weeks 1 + 2  
Concise Overview of Things Learned:

Technical Area:

- Count Vectorizer
- TF-IDF
- Introduction to LSTM and RNN Models

Tools:

- sklearn API
- Pytorch

Soft Skills:

- Presentation over Digital Medium

Three Achievement Highlights:

1. Use the CountVectorizer and Logistic Regression on the Amazon dataset.
2. Build TF-IDF Logistic Regression model used on the Amazon dataset which yielded an accuracy level of 74%
3. Present to my team about the Count Vectorizer model and present Team Amazon findings with all the models run.

Meetings Attended:

- Updates on Webscraping 6/8
- Data Cleaning + Last MInute Webscraping Problems 6/9
- Review and Refine our Data for ML 6/12
- Embedding Techniques + ML Algorithms 6/16
- Embedding Techniques + ML Algorithms 6/18
- Week 5 Planning + Team Building 6/19
- Presentation of Week 3 Teams Work 6/22
- Discuss Data Merging and Bert vs TF-IDF 6/24

Tasks Completed:

- I was able to learn about the count vectorizer word embedding and make some initial models to see how it functions.
- I then created a presentation and ran my team through what I learned, most notably that count vectorizer is a lite version of TF-IDF word embedding which should generally be used over count vectorizer.
- Some hurdles I faced this week were that I had a lot of work from my courses in school as it was the end of one of my summer classes. Initially, I was meant to present my word embedding on Monday 6/15 but thankfully, Rohit and Sara allowed me to present on 6/16 instead.
- I also was able to research and learn more about LSTM and BERT models which function differently from the previous models that we’ve discussed. I look forward to working with this through this next portion of the project.

Goals for Next Week:

- Work on the TF-IDF and Logistic Regression model on the combined dataset from the Flowster and Amazon teams.
- Also play around with BERT models to learn the difference between this method and previous.

---

<div class="post-metadata">

**Author:** ![mutaz\_ahmed](https://d2nat51eddc6b4.cloudfront.net/user_avatar/stemaway.com/mutaz_ahmed/32/1688_2.png) [@mutaz\_ahmed](https://stemaway.com/u/mutaz_ahmed)\
**Post date:** [July 9, 2020, 5:52pm UTC](https://stemaway.com/t/mutaz-ahmed-machine-learning-pathway/4049/2 "2020-07-09T17:52:10Z")

</div>

Concise Overview of Things Learned:

Technical Area:

- TF-IDF
- Logistic Regression
- Hypertuning Parameters

Tools:

- sklearn API
- gridsearchcv

Soft Skills:

- Presentation over Digital Medium

Three Achievement Highlights:

1. Built TF-IDF & Logistic Regression model to be used on the non-augmented merged dataset
2. Built TF-IDF & Logistic Regression model to be used on the augmented merged dataset.
3. Hypertuned the C parameter and the max\_iter parameter using gridsearchcv through SKLearn

Meetings Attended:

- Training the models results 6/29
- Updates with Evaluating the Models 7/01
- Preparing the Results Presentation 7/06

Tasks Completed:

- Built TF-IDF & Logistic Regression classifiers that ran on the merged dataset and the augmented dataset.
- With both models, I used hypertuning on the C and max iteration parameter of logistic regression to try and optimize the results
- I also contributed to the final presentation with the parts I completed for the project.
- One hurdle overcome was that hypertuning took a very long time train, depending on how much I varied the parameters, it could take hours. The google colab runtime kept disconnecting before the model would finish training due to inactivity. In order to solve this, I found a script online that, when pasted into the browser console, would imitate the user clicking on page every couple minutes, thus allowing the program to continue running.
