Machine Learning & Tensorflow - Google Cloud Approach

Meet Datalab and Jupyter in Google Cloud for interactive notebook execution in a browser, with options to clone notebooks, save changes to git, and connect via remote or local environments.

Discover how machine learning learns from data to infer patterns, classify inputs like emails and images, and compare traditional versus presentational approaches.

Machine learning uses feature vectors to map inputs to labels, while deep learning builds neural networks with layered neurons that automatically learn features for image classification.

Learn to perform simple math in TensorFlow by building a computation graph, executing it with a session, and visualizing the graph with TensorBoard.

Explore how TensorFlow uses an open source data-flow graph with computational nodes, edges, and tensors to power machine learning, training through feedback and iterative graph computations.

Explore how a graph organizes operations, run parallel clusters, and represent data with tensors, shapes, and bracket notation, defining the data type by its elements.

Explore tensors and tensor operations in a hands-on lab, building simple one-dimensional X and Y, performing products, and using reduce mean to compute averages.

Introduce simple linear regression, modeling y as a function of x with a least-squares best-fit line learned from training data. Examine how the slope, intercept, and error drive updates.

Train a simple linear regression model in TensorFlow to fit a line by optimizing w and b with mean squared error and gradient descent, then predict new x values.

Learn how placeholders and variables feed x and y into a regression model to fit a line by training with past data and handling runtime inputs.

This lab demonstrates using placeholders for inputs in a math graph, feeding x and y values through a feed dictionary, and executing a session to compute x divided by y.

Explore managing variables in a machine learning workflow by initializing and updating them inside a session, using a for loop to increment a multiplier and observe final results.

Explore how TensorFlow powers image processing with convolutional neural networks, representing images as tensors and extracting features like edges and shape for cat or dog classification during training and testing.

Upload images, read and decode them, and resize to a chosen shape; process with a TensorFlow session and queue in a multi-threaded workflow, then visualize results in tensorboard.

Explore how images are represented as tensors, with grayscale as a single channel and color as rgb channels, and how height, width, and channels form 3-d shapes.

Explore image preprocessing techniques in TensorFlow: decode and resize images, apply central crop and flips, and stack results into a single tensor for model input.

Explore the MNIST digit dataset, with 28x28 grayscale images and one-hot labels, and apply the k-nearest neighbors algorithm using distance metrics like Euclidean, to recognize handwritten digits.

Explore distance measures, focusing on the L1 (Manhattan) distance for classifying MNIST-like images with 784-pixel vectors, and compare it to Euclidean distance via a k-nearest neighbor approach.

Explore real time neural networks, combining linear transformations and nonlinear activation to learn regression and xor tasks, with backpropagation, training, weights and biases, and multi-layer architecture.

Explore linear regression with gradient descent, using data frames to compute returns, and optimize mean squared error by determining weights and bias for best fit predictions.

Explore logistic regression, predicting a probability between 0 and 1 for binary outcomes via the logit function. Compare with linear regression and apply a threshold for binary classification.

Explore softmax activation for multi-class probability outputs and cross-entropy cost for training estimators. Learn logistic regression basics, feature and weight shapes, one-hot encoding, and estimator-driven training with epochs and optimizers.

Load NYC taxi trip data, visualize daily trips, and plot demand against days and weather; establish a baseline and test hypotheses for a predictive modeling approach.

Explore building a regression model with TensorFlow on a small dataset, including data preprocessing, one-hot encoding considerations for categorical features, scaling, and evaluating with training and test splits.

Create a Google Cloud VM instance via the web console, naming it and selecting a region and machine type. Configure firewall and SSH access, then connect and install Python.

Edit a Google Cloud Compute Engine VM instance by stopping it to change machine type, adjust memory, add disks, and set access scopes; consider migration during maintenance.

Create and manage Google compute engine VM instances from the command line, using labels, default zones, and scripted commands to deploy multiple instances and enable SSH access.