Build and Deploy Machine Learning App in Cloud with Python

Explore how images are composed of pixels, grayscale ranges from 0 to 255, and RGB color channels combine to form colors using hue, saturation, and value.

Learn to read images with skimage, load and display an RGB image using matplotlib, and understand image arrays and color channels (red, green, blue) in the reading process.

Split the rgb array into its red, green, and blue channels, display the red, green, and blue matrices, and show how their combination reconstructs the original color image.

Convert a color image to grayscale with a Python library by combining red, green, and blue channels into a 2d grayscale array of 0 to 1, and save the result.

Learn to read a grayscale image, analyze its intensity with a 0–1 scale and 255-bin histogram, and understand image dynamics for future histogram equalization.

Learn how to resize an image to different shapes using transform.resize, setting output shapes like 200x300 or 800x1200 and adjusting from the original 400x600.

Prepare the dataset for image classification by understanding image size and shape, labeling data, and building a 20-class model in Python.

Explore the dataset structure with 20 class folders and 80 by 80 pixel images, showing partially pre-processed data. Learn how to load images and perform Python data preparation.

Read all images from the dataset and label them by extracting the folder names with a regular expression. Map images to lowercase labels and verify the total image count.

Read all images into an array of 2057 images with 80 by 80 by 3 dimensions, store in a data and labels dictionary, and pickle it for later use.

visualize label distributions and sample images to inspect dataset balance and class representations, using counter, bar plots, and dynamic subplots to display multiple classes.

Extract image features with histogram of gradients (hog) to improve model performance. Convert to grayscale, tune orientations and blocks, and build a hog-based pipeline.

This lecture builds an rgb to gray transformer using a base estimate and transformer mixin, applying fit and transform to single or multiple images to produce grayscale outputs.

Explore building a hog transformer to extract histogram oriented gradients from grayscale images, transforming image data into robust features for machine learning.

Train a stochastic gradient descent classifier using a pipeline that grayscales data, extracts features, and scales them, with grid search and cross-validation to tune learning rate and loss.

Test your model with a classification report on the test data, report accuracy, precision, and recall, and discuss multiclass performance and hyperparameter tuning using group search method.

Create a pipeline model for image classification with grayscale, transformer, scaling, and sgd classifier. Train and evaluate with a classification report; grid search tunes hyperparameters in the next lecture.

The lecture demonstrates grid search cross-validation to tune hyperparameters for a pipeline model, systematically enumerating orientation, pixels, loss type, and learning rate to maximize accuracy.

Build an end-to-end image pipeline that processes 80x80 inputs, applies transformations, trains a parameter-tuned model with grid search, and saves the pipeline and scaler with pickle for deployment.

Create a pipeline to generate predictions by loading a classifier, preprocessing images (resize to 80x80, grayscale, hog features), scaling with a saved scaler, and predicting in a Flask web app.

Compute class decision values, convert to probabilities with softmax, visualize results, and extract the top five predictions for a machine learning pipeline.

Install Visual Studio Code from the official site by downloading the stable version and set up extensions like autocomplete, Anaconda, Django, Flask, and hatchetman boilerplate for Python ML app development.

Set up a project template, organize static assets, and integrate bootstrap and jquery to build a web page ready for deployment on a cloud platform.

Learn to add a navigation bar and footer to a Flask web app, using a container for layout and live-browser previews for image processing pages.

Extend Django templates with a base layout and blocks to create reusable templates, override the body section, and render and test an upload page in the browser.

Create an image upload feature using a multipart http request, with an upload button and a file input named 'image', sending data to the flash gap for processing.

Learn how to style an html page with css by adjusting alignment, background color, padding, and text color, and apply styles to images and buttons for an image classification webapp.

Load data and models into a Flask app, apply a pipeline model with transform and scale steps to an uploaded image, and output the top five classification values.

Learn to dynamically set an image's width and calculate its height to preserve a constant aspect ratio, by reading image data and calibrating the height for a chosen width.

Style the output by applying HTML styling: set a background color, padding, and aligned content, and format a full-width table with border-collapse and a solid 1px black border.

Explore implementing robust error handling in a machine learning web app by configuring 404, 405, and 500 handlers, validating file extensions, and rendering user-friendly error templates.

Create an about page using templates and HTML, and organize content with containers and classes. Also prepare for deploying the webapp to the cloud next.

Prepare the Python web app dependencies by creating a requirements.txt, list libraries such as Flask, pandas, and SciPy with their versions, and install them using python -r requirements.txt.

Learn to prepare and upload a zipped Flask app to PythonAnywhere, use the bash console to unzip the package, verify python versions, and locate files, notebooks, and requirements for deployment.

Deploy a Flask web app to the cloud, configure the web app with the framework, upload your files, test with sample images, and access the app from anywhere.

Identify common deployment errors caused by environment settings and Python version mismatches; inspect logs, install the proper Python version (3.8), and address cache and permissions.