A 2-day Seattle Trip

 Seattle Downtown Sightseeing - 2 days

13th Sept 2023

Morning

- Pike's Place: One of the first few places that would show up if you search for Seattle sightseeing. This is situated next to the waterfront and you will find fresh fruits and vegetables, a fish market, restaurants, and a lot of souvenirs. 

Don't forget to try the Russian Bakery Piroshky Piroshky. I ate a Smoked Salmon pate there. 

- First Starbucks Coffee Shop: It has retained the old Starbucks logo but from what I have heard the menu is the same as any other Starbucks. Always crowded! 

- The Gum Wall: It was once named the second germiest tourist attraction in the world. All you gotta do is stick the chewing gum on the wall. God knows who decided to make this a tourist place 😂

- Wings over Washington: This is a 4D ride that takes you virtually through the different state parks in Washington State. A unique experience as if you are seated in a helicopter, this is definitely worth visiting. 

Evening 

- The Spheres by Amazon 

- Kerry Park: You get nice city views from this park

- Sculpture Park: A serene place with multiple sculptures surrounded by the waters to spend your evening

 

14th Sept 2023

Morning 

- Space Needle: This is an observatory deck that I skipped but clicked. 

- Chihuly Glass Garden: Located beside the Space Needle, it is an art gallery consisting of glass artifacts that shouldn't be missed at all. Mesmerizing to look at the different glass designs. There are some pieces for sale as well but be ready to shell out $1000 or more. 

Evening: 

Snoqualmie Falls - This is located at a half-hour drive from downtown. Waterfalls are always a treat to watch and the abundance of trees next to the falls adds to the beauty of the landscape. There is also a hike here. 

Bonus: 

New York Trip (7th Sept - 10th Sept 2023) in pictures

Day 1 - 7th Sept 2023

Niagara Falls & Maid of the Mist boat ride

Day 2 - Reached NYC at 4pm 

- Empire State Building 

- Times Square 

Day 3 

- Statue of Liberty 

- Financial District: Charging Bull, NY Stock Exchange

- Lunch at Halal Cart 

- 9/11 Memorial 

- Central Park: Sheep Meadows, Bethesda Terrace 

- Brooklyn Bridge 

- Dinner at Joe's Pizza

If one is in New York, they should also definitely try out the Bagels. We visited a place called Leo's Bagels. There are plenty of options to choose from and one wouldn't complain about the portion size. 

Road Trip Southern to Northern California in Pictures

We had planned to travel along the CA -1 (Pacific Coast Highway). However, parts of the highway were closed but we still had loads of fun visiting these places. It was extremely hot in California a couple of days before the road trip but thankfully the weather God decided to be generous and we were blessed with very pleasant weather during the road trip. 

Here are the stops and the stunning views along our road trip. 

1. Morro Bay - Finally explored the Panoramic mode of the phone because of the wide landscape 😆

2. Paso Robles - Street photos: A wonderful stop mid-journey. The place is famous for its wineries and olive groves. 

3. Carmel by the Sea - A must-stop place. Simply walking on the streets of this place is fun. 

We then did a tour of the 17-mile drive which surprisingly has 17 spots. It is a scenic drive through the Pacific coastline which lies in the Monterey Peninsula of California. The next few photos are from the 17-mile drive. 

4. The Landscape

5. Pebble Beach and the Restless Sea

6. Lone Cypress Tree

7. Pescadero Point - Ghost Tree

And a few other snaps: 

Medium Articles

As medium.com has an easier interface, I have decided to switch to Medium for articles related to technology. 

I will post the links to the medium articles here:

Deploying a Machine Learning Model- Getting Started

Building Neural Networks like Lego blocks — Using Keras

ML Orchestration — Kubeflow

Getting Indian Stock Prices Using Python

Installing Python Virtual Environments

Deploying a Machine Learning Model: Part 2 — Local Deployment

An encounter with a wise man

I recently visited an ISKCON temple along with my friend and I had a very interesting encounter with a wise man. We were having the Prasad after the darshan when I spotted a person who seemed to radiate positive energy. Generally, I don’t talk to strangers but an interaction with this person seemed imminent.

‘How did you find this temple?’ was the first question I asked. ‘It was good but there are certain things that I can’t comprehend. First of all, there are signboards all over the place which say that chanting the Lord’s name for so and so times will help you in overcoming the challenges of the Kalyug. I’m perplexed looking at these signboards. I understand that chanting the Lord’s name will help in building your courage and faith but it is an individual’s responsibility to go face to face with the challenges and come out at the top. That’s the essence of Karma. I felt that the element of faith could have been better explained to the masses. What do you think about it?’

I replied ‘I couldn’t agree more with what you just said. I think they have put up these boards as people look for an idol, for a god that can come to their rescue when the times are tough. In these situations, it becomes difficult to grasp the real meaning of faith and courage. So I would say that these boards may be a source of strength for a lot of people and that’s why these boards are up.’

The man seemed to like my answer and he continued ‘I like how you are trying to balance the scenario.’

I then moved on to the topic of the temple’s aesthetics. I said ‘No doubt the temple is beautiful but a lot of money has been invested into building this place. I felt the money should have been invested elsewhere maybe building educational institutions or healthcare facilities.’

The man replied ‘You seem to have a kind intention of trying to help people with monetary means. However, what you’re thinking is an ideal scenario, and the world as you must be knowing is imperfect. While people should be spending some time in a peaceful environment getting away from their countless thoughts they fail to do us. So the temple provides an abode for the people to pray in silence moving away from their thoughts and focusing on the lord. This is true for people of all religions. But I do agree that temples should be simplistic and not too glitzy as that defeats the purpose of humble living.’ I was amazed by this insight.

He then moved to a thought that I was holding in my mind. ‘I think we both will agree on the point that it is unnecessary to sell Lord’s posters and idols at an exorbitant price. There are many places in this temple where we read about how a materialistic lifestyle cannot lead to a happy life and then there are posters so heavily priced that force you to have materialistic gains. That seems to contradict the very philosophy of moving from a materialistic to a spiritual lifestyle.’

It was turning out to be a fruitful conversation when I asked this person his name.

He said ‘My name is Krishna…’

Guidelines for a leadership position

January 2020: I was promoted to the role of an Apprentice Leader at Mu Sigma. Apprentice leader can be thought of as a Data Science Manager where we have to continuously strive to improve the quality of the team that will be conducive for its growth. Transitioning into a manager role was certainly challenging. In addition to solving technical problems, there were new responsibilities such as collaborating with multiple people, ensuring that your team is motivated and inspired.

I have listed down a few guidelines (and not rules) that I felt are important to be successful in a leadership position.  

1. Willingness to take up challenging tasks: I will be doing a lot of work that I haven’t done before and I will be pushed out of the comfort zone. I need to be prepared for this. In order to get results for these tasks, the points below will be of utmost importance
2. Time management: The importance of time management can never be stated enough. When there are busy days, having a to-do list and a schedule will be very effective. Also, it is important to be in the present and not think about other issues at that point in time. Context switching is a skill that comes in handy in management
3. Not getting affected by environment and people: Be ready to listen to opinions that are not aligned with my opinions. Be ready to be criticized and not get pissed off when people show a lack of understanding. In such cases, if I feel that I have a strong point that can show a different perspective based on my observations, then make sure it is being articulated in a calm and composed manner without insulting or criticizing the other person. Also getting angry is natural but wasting a considerable amount of time because of the anger is not worth it
4. Be strong and decisive: The ability to take timely and tough decisions are indispensable in a leader’s toolkit. Additionally, I need to be impartial and unbiased while making decisions. It is difficult but it is what the role demands
5. Be a powerful narrator: Many a time it will be necessary to promote the work being done or to present an impactful project. In such cases, it is essential for me to have the necessary storytelling skills with a powerful narrative. The message has to be loud and clear when discussing something critical
6. Be persistent in your initiatives: No matter how lofty any idea is, it will not survive the test of the time if the leader is not being persistent with it. Does that mean the leader should never change any plan? No that’s not true. If the data shows that the plan is not working out then the leader should be the first one to quickly change the plan and take learnings from the failure
7. Be willing to change beliefs and personality depending on the circumstances: I have faced this issue many times owing to my previous role. Now that I have been promoted I should grab this opportunity and hone my leadership skills
8. Understand when you are getting stressed and relax: Generally stress leads to stiffening of the body. It is easy to be observe that. In such cases, taking a deep breath relaxes our muscles. The brain -- by observing that the tension no longer exists in the muscles -- will be free of stress and will lead to positive reinforcement of a good mood
9. Be ready to seek help and talk to people: Talking to people gives us a new perspective and also leads to an improvement in our networking skills. Asking for help when stuck in a problem is also equally important
10. Consistency: This is a crucial dimension as I have seen that consistency leads to incredible results. Consistency in sleeping time especially on Weekdays is key to get a good rest and waking up fresh. Consistency in going to gym builds up the momentum and develops a good physique. Consistency in preparing breakfast leads to a healthy diet and good digestion. Consistency in planning the day right after going to the office gives a direction for the rest of the day

While these 10 principles are being followed, it is also necessary to take a break and reward yourself.

Cross Validation - Model checking, not model building

1st Update on 11th August 2020

2nd Update on 4th August 2023

3rd Update on 25th September 2023

Cross-Validation 

Cross-validation is a popular technique that is used for model evaluation in machine learning. If you skim through any Kaggle Kernel then you're bound to find Cross-validation in the script. It allows us to evaluate how our model will work when faced with unseen data and extends the concept of train-test datasets. It is also used to counter overfitting by examining how well our model generalizes.

CV has 2 main objectives: 
1. Give a good estimate of how our model generalizes to unseen data
2. Compare the performance between different models or a combination of hyperparameters 

I had this question about cross-validation. Suppose we do 5 fold CV, we are splitting our dataset into 5 portions and then training and testing on each portion. Then how are we supposed to choose which of these models is the best?

A similar question was asked on StackExchange which cleared my concern.

Also, this link from Machine Learning Mastery was helpful.

An example of choosing the best type of model. Let's consider a classification problem and we want to try Logistic Regression and Random Forest. Now, in this case, we will compare the CV error of both methods and choose the one that minimizes the error. 

Suppose Logistic Regression is a winner here. We will thus discard all the models that were built during the CV and train the model on the entire dataset. This model will then be utilized for predictions.

An extension to the above problem: We considered Logistic Regression and Random Forest in the above scenario but the question of hyperparameter tuning will now arise.
In logistic regression, we have L1, L2 regularization and in the random forest, we can vary the number of trees, maximum number of features etc. Hence, we are not just trying 2 models but a lot of variations amongst these models as well. Scikit Learn's GridSearchCV can be used here as it will iterate through all the variations and find the optimal model with its hyperparameters.
Now that we are aware of the best model along with its hyperparameters, we can train the model on the entire dataset.

Cross-validation Pseudocode:

Here we divide the training dataset into k-folds and k-1 folds are used for training and the remaining fold for validation

1. For 1 to N hyperparameter combinations:

    i. For 1 to k folds: #(The step repeats holdout approach k times)

        Train the model on k-1 folds and compute the validation error.

    ii. Calculate the average validation error

2. Choose the hyperparameters with the lowest average validation error and evaluate the model on test error

Here is the link to an example jupyter notebook.

Holdout validation:

We understood the concept of cross-validation but there is yet another strategy that can be adopted for choosing the best model or the model with the best hyperparameters. The strategy is called "holdout validation". 

Here, the dataset is divided into 2 sets: training and testing. However, the training dataset is further divided into training and validation sets. Thus, we would train our model on the training data and then evaluate its performance on the validation dataset. We could then compare the validation metrics between different models or if we want to choose the best hyperparameters. 

Once, the best model is selected we can then train it on the entire dataset and compute the test dataset metrics to understand how well our model generalizes. The con of this method is that we would choose only a small portion like 10% of the training data as our validation data and our decision would be based on this validation metric. 

Holdout Validation Pseudocode:

1. For 1 to N hyperparameter combinations:

    Train the model on the training dataset and compute the error on validation data.

2. Choose the hyperparameters that generalize the best and evaluate the model on test error

Here is a great link that explains the difference between holdout validation and k-fold cross-validation.