Deep Learning Mastery

Explore the shape of arrays and learn how to reshape and flatten them in numpy, converting 1d arrays to 2d or 3d and manipulating dimensions.

Learn how to access elements in one-dimensional arrays using zero-based indices. Explore two-dimensional array indexing with row and column positions, including examples like a[0], a[2], and B[1,4].

Learn to iterate in Python with for loops over 1d and 2d arrays, print elements, and use nested loops to access each scalar value.

Master Python NumPy slicing by learning start and end indices, steps, and zero-based access for 1D and 2D arrays, with practical examples and Jupyter notebook demonstrations.

Learn pandas, a Python library for data analysis, storing datasets in a data frame and performing cleaning, exploring, and manipulating data.

Read csv files with pandas read_csv to store data in a dataframe, then print and inspect the full dataset, ensuring the csv is in the same directory as the notebook.

Explore pandas to analyze data frames by using head and tail to view data, info and describe for basic statistics, and correlate features to reveal relationships.

Explore different types of plots in matplotlib, including line, bar, scatter, pie, and histogram, using numpy arrays and plotting commands for data analysis and visualization.

Learn to install and import seaborn, then build box plots, distribution plots, and a rich plot using built-in datasets to explore data, outliers, and feature relationships for machine learning.

Explore supervised machine learning, where models learn from labeled training data to map inputs to outputs, using a lifecycle of training and testing with regression and classification tasks.

Explore unsupervised machine learning, where models train on unlabeled data to discover hidden patterns, group data via clustering and association, including market basket analysis, and reveal underlying structure without supervision.

Explore the machine learning lifecycle from data gathering and preprocessing to model training, evaluation, and deployment, emphasizing data quality, integration, and metrics like accuracy and precision.

Learn to identify and handle missing values in data, understand why values go missing, and apply mean, mode, or simple imputer techniques to fill numeric and categorical features.

Understand feature scaling in data preprocessing, including normalization and standardization, with min max scaling and practical notes on when to scale for neural networks versus tree-based models.

Explore feature encoding techniques for categorical data, including nominal and ordinal encoding, one hot and dummy encoding, and label encoding, with practical pandas examples.

Evaluate models with confusion matrix, accuracy, precision, and recall; address class imbalance and use mean absolute error or mean squared error to improve generalization before production.

Explore activation functions in neural networks, including sigmoid, tanh, and ReLU variants, and understand how they enable non-linear transformations, backpropagation, and appropriate output bounds for binary and multiclass tasks.

Explore optimizers in neural networks, including gradient descent, stochastic gradient descent, mini-batch gradient descent, and Adam, to efficiently reduce loss and update weights and learning rates.

Train an artificial neural network to predict gold prices from S&P 500, silver, euro/usd, and oil, using scaling and a four-feature regression model, achieving an r2 score around 73%.

Develop a binary diabetes prediction model with an artificial neural network in TensorFlow Keras, from loading the diabetes dataset to training, evaluating, and achieving initial accuracy.

Explore convolutional neural networks for image data, learn how filters and receptive fields extract features, apply pooling and strides, and build CNNs with TensorFlow and Keras for vision tasks.

Implement a convolutional neural network with Keras and TensorFlow, building a sequential model with conv2d, max pooling, flatten, and dense layers, and train on ten-class image dataset with 20 epochs.

Explore recurrent neural networks, which feed outputs from the previous step as inputs to the current step to remember past data, enabling tasks like predicting the next word.