-
Today I studied the concept of GAN(Generative Adversarial Network).
-
Implemented DCGAN(Deep Convolutional Generative Adversarial Network) on the MNIST dataset to generate handwritten digits.
-
I Would love to explore the applications of GAN in the upcoming days.
-
Link to the notebook. Click here.
Deep Convolutional Generative Adversarial Network (Tensorflow doc)
What is Generative Adversarial Networks GAN?
-
Studied various regularization techniques in order to handle overfitting.
-
Completed Week 1 assignment of the Andrew Ng Deep Learning course.
-
Explored various optimization algorithms such as Gradient descent with momentum, RMSprop and Adam.
-
Completed Week 2 assignment of the same course.
-
Course link : Click here
-
Implemented DCGAN(Deep Convolutional Generative Adversarial Network) on the face dataset to generate human faces.
-
The image generated was not upto the mark and would need some modification.
-
The dataset is available Here
-
Link to the notebook is available here
-
Learnt about Machine Learning structural strategies used in the industry.
-
Read about terms like Avoidable bias and variance on the dataset and about Bayes error.
-
Read about Transfer Learning, Multi-task learning and end-to-end deep learning.
-
Completed the course 3 of the deep learning series by Andrew Ng.
-
Today I studied about the various clustering algorithms used in ML.
-
The list includes : K-means, Mean-shift, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Gaussian Mixture Models (GMM) and Agglomerative Hierarchical Clustering.
-
Analysed the pros and cons of each algorithm and the applications of each of them.
-
Looking forward to implement them in the coming days.
-
Today I started course 4 of the Deep Learning Series by deeplearning.ai
-
Week 1 turned out to be a quick recap of Convolution Neural Network (CNN).
-
The topics covered todaytujhe bhula diya included padding, strides, convolutions and MaxPooling.
-
The course is available Here.
-
Today I worked on the conversion of black images to images of color.
-
Used OpenCV and Deep Learning to implement the program.
-
The model used was pre-trained on the Caffe deep learning framework on ImageNet dataset.
-
You can refer to the paper here and to the documentation here
-
The results were plausible.
-
Today I dived deeper into the background of CNN.
-
I wrote the code for it to understand what happens behind the scenes.
-
It's always good to know about the theory and code of the pre-defined functions used directly by us.
-
I implemented the model on the 'signs' dataset.
-
Today I started learning about the basics of the PyTorch deep learning framework.
-
Looked up the differences between Tensorflow and PyTorch.
-
Implemented the framework on the University of California car dataset.
-
Here's the notebook for it.
-
Today I started Week 2 of the fourth course in the Deep learning specialization.
-
Various architectures like ResNet, AlexNet and Inception were discussed in detail.
-
Read about the effectiveness of 1x1 convolution.
-
Would soon read the research papers on these architectures.
-
Today I implemented the CIFAR 10 dataset using Pytorch.
-
The accuracy was not upto the mark since my primary focus was on understanding the framework.
-
Would tweak the hyperparameters and increase the number of epochs to achieve higher accuracy.
-
Would implement Pytorch on different datasets as well in the coming days.
-
You can find the notebook here.
-
Today I studied the required mathematics for machine learning.
-
The topics included Linear Algebra, Multivariate Calculus, Probability and Calculus.
-
The YouTube video for it is available here.
-
Today I worked on the analysis of Mall Customer Segmentation data on Kaggle.
-
The analysis involved the comparison between parameters like Customer age, salary and gender.
-
K-means algorithm was used to form a cluster of customers on the basis of their shopping traits.
-
The notebook is available here.
-
Today I read the theory of time series analysis and its applications.
-
I was fascinated to know that time series analysis has a different approach as compared to conventional ML algorithms.
-
The applications of time series really intrigued me and would implement it in the coming days.
-
Relevant links.
7 Ways Time Series Forecasting Differs from Machine Learning
-
Completed the assignment 1 for SHALA2020 course.
-
The assignment was an introduction to important libraries like Numpy, Pandas and Matplotlib.
-
The notebook is available here
-
Revised statistic concepts assigned under the pre-work category on SHALA course.
-
Consisted of topics like measuring central tendency, histograms and statistic fundamentals.
-
Eventually gave a quiz on the related topic.
-
Relevant links.
-
Completed assignment 2 of the IITB Shala course.
-
Plotted and analysed data using Histograms, boxplots and pie chart.
-
The notebook can be accessed here.
-
Predicted the airline traffic for the future(3 years) using time series analysis using fbProphet.
-
Used Markov Chain Monte Carlo method(MCMC) to generate forecasts.
-
The model shows both the seasonality and trend in the data.
-
You can check out the notebook here.
-
Predicted the milk production for the future(3 years) using time series analysis using fbProphet.
-
Used Markov Chain Monte Carlo method(MCMC) to generate forecasts.
-
The model shows both the seasonality and trend in the data.
-
You can check out the notebook here.
-
Read the theory of different types of graphs and their uses.
-
Studied the differences between the types of graphs and charts.
-
Learnt several visualization practices.
-
Relevant links.
-
Completed assignment 2 of the IITB Shala course.
-
The assignment consisted of different kind of charts and graphs that can be made using libraries like Pandas, Matplotlib and Seaborn.
-
Explored various other visualization techniques as well.
-
You can access the assignment here.
-
Read a lot of intermediate statistics concepts.
-
Topics included Maximum likelihood estimation, sufficient statistics, null hypothesis testing, t-test and Wilcoxon rank test.
-
Would implement these concepts in the future.
Relevant Links:
-
Completed assignment 4 of the SHALA IITB course.
-
With this the module 1 (Data Science) of the course has been completed.
-
Computed the likelihood and log likelihood from samples that were drawn from an exponential distribution.
-
Performed a two sample t-test from samples of unknown distributions and found the critical value.
-
Notebook is available here.
-
Started week 3 of the CNN course by deeplearning.ai
-
Read about object localization, object and landmark detection and non-max suppression.
-
Studied the YOLO algorithm and implemented it in the week's assignment.
-
Was able to complete the assignment and detect cars.
-
Started Week 4 of the CNN deeplearning.ai course
-
The topic for the week was art generation with neural style transfer.
-
Understood the 2 types of cost functions i.e. Content cost function and Style cost function.
-
Completed the assignment for art generation.
-
Read about one-shot learning and its application in face recognition.
-
Understood the concepts of Siamese network and triplet loss.
-
Implemented face recognition and verification in the assignment.
-
With this completed the Course 4 of deeplearning.ai specialization.
-
Read about Decision trees and Random Forest Regressor as well as Classifier.
-
Gave a quiz on the topic and implemented the concepts in the SHALA ML assignment 1.
-
Predicted the Attrition of the employees of a company using classifier.
-
Made a classifier using Logistic Regression to predict the survival of a passenger in the Titanic dataset.
-
Reached 79.3% accuracy, plotted the roc curve and calculated the roc_auc_score.
-
Found out that Jack Dawson couldn't survive whereas Rose made it. Sigh!
-
The notebook can be found here.
-
Made several classifiers to predict passenger survival on the Titanic.
-
Applied Logistic Regression, Decison Trees, Random Forest, Gradient Boosting and XGBoost algorithms.
-
Corrected the anomalies of the logistic regression analyis performed the previous day.