Here, we apply Machine learning algorithms to build predictive model and build end-to-end application:
The goal of repositories amitkr2410/MachineLearning and amitkr2410/MachineLearning-Deployment is to explore all stages from preparation to deployment of machine learning models and build end-to-end applications. In this repository, we present machine learning models for numerical data, NLP and computer vision tasks using Sci-Kit Learn and PyTorch library.
The goal of this project is to build CT scan tumor detection app using self-attention module and compare the performance with traditional VGG16 architecture. We deploy the application on Google Cloud Run. To access the web app files, click here:
PyTorch based WebApp for BrainTumor detection on Google Cloud Run
The goal of this project is to build CT scan tumor detection app using self-attention module and compare the performance with traditional VGG16 architecture. We deploy the application on AWS server using AWS ECR and AWS Lambda. To access the web app files, click here:
PyTorch based WebApp for BrainTumor detection on AWS Cloud
The goal of this project is to build CT scan tumor detection app using self-attention module and compare the performance with traditional VGG16 architecture. We deploy the application on Google Kubernetes Engine . To access the web app files, click here:
PyTorch based WebApp for BrainTumor detection on Google Kubernetes
**Personal Homepage: Amit's Personal HomePage
In below, we present a typical layout of data pipelines and ML experimentation for prototyping a model.
The goal of this project is to build CT scan tumor detection app using
self-attention module with positional encoding and compare the performance with traditional
VGG16 architecture. We deploy the application on Google and AWS cloud.
For more details, refer to the README.md in thte project directory.
Introduction to SVM (Supervised machine learning technique)
Useful for classification, Regression, Outlier/Anomaly detection
Relevant Library to Import
Modeling Cancer data to predict whether a cell sample is benign or malignant.
To view the project click on "main_svm.ipynb"
In this project, we explore how to fine-tune ROBERTa transformer model.
We fine tune on custom paragraphs for Q/A taks.
View projects "fine_tune_RoBERT.ipynb" and "main.ipynb"
In this implementation, we use single-head self-attention layer with positional
encoding and construct a custom transformer moodel. We train the model on
sentimental analysis data where the task is to classify the sentiment of text
into positive, neutral and negative.
View project file "TransformerScratch.ipynb"
Natural language processing to classify a resume into a job category
Read reasumes in pdf format using pdf2image ocr tool and convert it to
PIL image, then use pytesseract.image_to_string to convert the image into strings
objects. For webscrapping, we used BeautifulSoup and Selenium driver.
Next, we use natural language processing tools such as NLTK stopwords, regex,
string replace to get important words from the string.
We also remove prefix and suffix using PorterStemmer in NLTK library.
Then, we use CountVectorizer method and TfidfVectorizer method from sci-kit learn
to create a independent feature vector.
We use explore NaiveBayes model to train the data set. Compute confusion matrix.
To view the project click on "Scrapping_resume.ipynb" and "NLP_modeling.ipynb"
Classify brain tumor 2D scanned images into tumorus or healthy
sample using Convolutional neural network.
We use keras library to create a model architecture and train the model using Adam algorithm.
The original images of brain are converted into a matrix data using
function "ImageDataGenerator" defined in keras library.
We study the accuracy at each epoch of the model training. In the end,
we test the model againt known data and compute precision score and confusion
matrix to determine the validity of the model.
To view the project click on "main_cnn.ipynb" or "main_cnn_resnet50.ipynb"
We apply Natural Language Processing (NLP) techniques to classify the sentiment
of the text message as positive, neutral, or negative using Python. The text
preprocessing is done using "stopwords" and "PorterStemmer"from NLTK corpus
library and then, "CountVectorizer" function is used to construct a 2d-matrix
from the text messages. The 2d-matrix is used to build the model for the
following different cases:
(a) Decision trees
(b) Random Forest method
(c) Logistic Regression model
(d) Support vector classification (SVC)
(e) xGBoost classifier
(f) Naive Bayes
We compute the accuracy, precision, f1-score (confusion matrix) to determine the
performance of the machine learning model.
To view the project click on "NLP_Sentiment.ipynb"
We apply statistical models:
(a) Decision trees
(b) Random Forest method
(c) Logistic Regression model
(d) Support vector regression (SVR)
(e) KNeighbors regression
(f) xGBoost regression
to study feature and corrleations present
in the Customer chrun data at a Bank. We convert the predicted probabilities into
binary group and compute the confusion matrix to determine the performance
of the machine learning model.
To view the project click on "CustomerChurn.ipynb"
We apply Natural Language Processing (NLP) techniques to classify a text message
as a spam or good email (ham) using Python. The text preprocessing is done using
"stopwords" from NLTK corpus library and then, "CountVectorizer" function is used
to construct a 2d-matrix from the text messages. The 2d-matrix is used to build
the model
for the following different cases:
(a) Decision trees
(b) Random Forest method
(c) Logistic Regression model
(d) Support vector classification (SVC)
(e) xGBoost classifier
We compute the confusion matrix to determine the performance
of the machine learning model.
To view the project click on "NLP_DetectSpamEmail.ipynb"
In this project, we do exploratry study of the data
Things to check: shape/size of data, nulls for missing values, types of coloumns
Use sci-kit library's SimpleImputer and OneHotEncoder to do preprocessing of the
data
Perform analysis to look for a trend in the data
Build Linear or polynomial regression model on the training data
To view the project click on "Project0_KaggleProject_ExploreData.ipynb"
or "Project1_PreprocessingAndRegression.ipynb" or "Project2_RegressionLinearPoly.ipynb"
We use python library called "BeautifulSoup" to do webscrapping.
In the end we use seaborn library to make a bar graph.
To view the project click on "WebScrappingBS4.ipynb"