Time Series Analysis and Forecasting with Python

Master numpy basics for time series analysis in python by importing numpy as np, creating a 2 by 3 array, and exploring it in Spyder's variable explorer.

Explore numpy arrays by creating two 2x2 arrays and performing matrix multiplication with the @ operator or np.dot, or element-wise multiplication with the * operator and np.multiply.

Learn element-wise addition and subtraction of NumPy arrays using operators and np.subtract, then explore broadcasting when adding scalars or differently shaped arrays, and use np.sum to total elements.

Master NumPy division by applying elementwise division and floor division on arrays, using NumPy divide and floor_divide with scalar or array divisors, and explore NumPy broadcasting and sqrt.

Learn to generate random data with NumPy, creating normal and uniform distributions, random arrays, zeros and ones, and practice filtering arrays with boolean logic.

Review mean, variance, standard deviation, and median in numpy, compute them for one- and two-dimensional arrays with axis, explore normal and uniform distributions, and preview pandas.

Explore pandas basics by creating series and dataframes, labeling with custom indices, filtering data, and accessing values and indices to prepare data for analysis.

Explore selecting data with pandas iloc and loc, distinguishing label-based versus integer-based indexing, using row and column positions, slices, and practical examples for extracting subsets.

Visualize monthly electricity consumption for two customers with Matplotlib, using colors, markers, labels, and a legend. Create side-by-side subplots to compare graphs clearly.

Learn to create scatter plots, histograms, and bar charts with matplotlib, customizing color, labels, titles, grid, and legends for clear data visualization.

Learn how to visualize box plots with Matplotlib, customize notched and rectangular forms, and set colors for box, whiskers, caps, and median to aid outlier detection.

Learn how to resample pandas data by changing its datetime index frequency and resolution, using a rule and mean aggregation to convert hourly data to daily or yearly.

Import key libraries, prepare a date-time indexed dataset, and develop the first ARIMA model on temperature data, using seasonal decomposition to inspect trend and seasonality.

Explore arima model development by using grid search with auto_arima to determine p, d, and q, compare models by aic, and confirm the best arima configuration for series forecasting.

Develop an arima model with order (1,1,2) on the training data, fit it, and generate level predictions. Compare forecasts to test data with matplotlib visuals and rmse.

Explore how the SARIMAX model extends ARIMA with seasonality and exogenous variables, define PDQ and seasonal length, and use grid search guided by decomposition.

Explore developing a SARIMAX model to forecast temperature using the same data as ARIMA, including exogenous variables, train/test splits, and model comparison.

Learn how to develop a sarimax model with exogenous variables using auto.arima, determine seasonality with seasonal_decompose, and set m=7 to forecast temperature data.

Explore the basics of deep learning by comparing neurons—dendrites, cell body, axon, and terminals—to a neural network with input, hidden, and output layers, weights, bias, and activation function.

Examine how a practical neural network learns: from input and hidden layers with selective connections, through the output's predicted value and backpropagation to minimize the cost function across all data.

Explore how gradient descent optimizes neural networks, comparing batch and stochastic approaches, tuning learning rates to converge to global or local minima, and applying backpropagation in a feedforward network.

Discover recurrent neural networks for time series analysis, featuring memory through hidden-layer feedback and back propagation through time (Bptt). Compare RNNs with feedforward nets and explore challenges.

Explore how long short-term memory (LSTM) networks overcome long-term dependencies in recurrent neural networks and enable robust time series forecasting with Python.

Develop a univariate LSTM workflow by loading solar radiation data, cleaning missing values, splitting training and test sets, and scaling with a min-max scaler to 0-1.

Create a 24-hour lookback window to build univariate x_train and y_train from the training set, forming batches for the LSTM model. Convert the lists to NumPy arrays for training.

Develop a univariate time series LSTM in Keras, building a four hidden layer network with dropout and a dense output, trained with Adam optimizer and mean squared error.

Develop a univariate LSTM for time series forecasting by forming lookback batches, testing with the same structure, and using a sliding window to generate successive future predictions.

Develop a multivariate LSTM model to forecast electricity consumption from humidity and temperature, using 24-hour rolling windows, data cleaning, scaling, and train-test preparation.

Develop a multivariate LSTM model in Python using Keras, with stacked LSTM layers and dropout, trained on Xtrain and Ytrain, compiled with Adam and mean squared error, saved for forecasting.

Explore testing a multivariate LSTM for time series by using a lookback window of multiple variables to predict the next target value and roll forward with predictions.

Develop a multivariate LSTM model using the last 24 hours as a 3-feature input to predict the next 48 hours, by building and testing batches from the scaled training set.