TigerForecast reference documentation¶
For an introduction to TigerForecast, start at the TigerForecast GitHub page.
tigerforecast package¶
Subpackages¶
tigerforecast.utils package¶
dataset_registry¶
unemployment([verbose]) |
Description: Checks if unemployment data exists, downloads if not. |
uci_indoor([verbose]) |
Description: Checks if uci_indoor data exists, downloads if not. |
sp500([verbose]) |
Description: Checks if S&P500 data exists, downloads if not. |
crypto([verbose]) |
Description: Checks if cryptocurrency data exists, downloads if not. |
enso(input_signals, include_month, …) |
Description: Transforms the ctrl_indices dataset into a format suitable for online learning. |
random¶
set_key([key]) |
Descripton: Fix global random key to ensure reproducibility of results. |
generate_key() |
Descripton: Generate random key. |
get_global_key() |
Descripton: Get current global random key. |
optimizers¶
optimizers.Optimizer([pred, loss, …]) |
Description: Core class for method optimizers |
optimizers.Adagrad([pred, loss, …]) |
Description: Ordinary Gradient Descent optimizer. |
optimizers.Adam([pred, loss, learning_rate, …]) |
Description: Ordinary Gradient Descent optimizer. |
optimizers.ONS([pred, loss, learning_rate, …]) |
Online newton step algorithm. |
optimizers.SGD([pred, loss, learning_rate, …]) |
Description: Stochastic Gradient Descent optimizer. |
optimizers.OGD([pred, loss, learning_rate, …]) |
Description: Ordinary Gradient Descent optimizer. |
optimizers.mse(y_pred, y_true) |
Description: mean-square-error loss :param y_pred: value predicted by method :param y_true: ground truth value :param eps: some scalar |
optimizers.cross_entropy(y_pred, y_true[, eps]) |
Description: cross entropy loss, y_pred is equivalent to logits and y_true to labels :param y_pred: value predicted by method :param y_true: ground truth value :param eps: some scalar |
boosting¶
boosting.SimpleBoost() |
Description: Implements the equivalent of an AR(p) method - predicts a linear combination of the previous p observed values in a time-series |
autotuning¶
autotuning.grid_search.GridSearch() |
Description: Implements the equivalent of an AR(p) method - predicts a linear combination of the previous p observed values in a time-series |
tigerforecast.problems package¶
custom¶
tigerforecast.problems.CustomProblem() |
Description: class for implementing algorithms with enforced modularity |
tigerforecast.problems.register_custom_problem(…) |
Description: global custom problem method |
tigerforecast.problems.MyProblem() |
Description: Make dataset into tigerforecast problem class. |
time_series¶
tigerforecast.problems.Problem() |
|
tigerforecast.problems.SP500() |
Description: Outputs the daily opening price of the S&P 500 stock market index from January 3, 1986 to June 29, 2018. |
tigerforecast.problems.UCI_Indoor() |
Description: Outputs various weather metrics from a UCI dataset from 13/3/2012 to 11/4/2012 |
tigerforecast.problems.ENSO() |
Description: Collection of monthly values of control indices useful for predicting La Nina/El Nino. |
tigerforecast.problems.Crypto() |
Description: Outputs the daily price of bitcoin from 2013-04-28 to 2018-02-10 |
tigerforecast.problems.Random() |
Description: A random sequence of scalar values taken from an i.i.d. |
tigerforecast.problems.ARMA() |
Description: Simulates an autoregressive moving-average time-series. |
tigerforecast.problems.Unemployment() |
Description: Monthly unemployment rate since 1948. |
tigerforecast.problems.LDS_TimeSeries() |
Description: Simulates a linear dynamical system. |
tigerforecast.problems.LSTM_TimeSeries() |
Description: Produces outputs from a randomly initialized recurrent neural network. |
tigerforecast.problems.RNN_TimeSeries() |
Description: Produces outputs from a randomly initialized recurrent neural network. |
tigerforecast.methods package¶
time_series¶
tigerforecast.methods.AutoRegressor() |
Description: Implements the equivalent of an AR(p) method - predicts a linear combination of the previous p observed values in a time-series |
tigerforecast.methods.LastValue() |
Description: Predicts the last value in the time series, i.e. |
tigerforecast.methods.PredictZero() |
Description: Predicts the next value in the time series to be 0, i.e. |
tigerforecast.methods.RNN() |
Description: Produces outputs from a randomly initialized recurrent neural network. |
tigerforecast.methods.LSTM() |
Description: Produces outputs from a randomly initialized LSTM neural network. |
tigerforecast.methods.LeastSquares() |
Description: Implements online least squares. |
tigerforecast.methods.WaveFiltering() |
Description: Predicts the last value in the time series, i.e. |
boosting¶
tigerforecast.methods.boosting.SimpleBoost() |
Description: Implements the equivalent of an AR(p) method - predicts a linear combination of the previous p observed values in a time-series |
tigerforecast.experiments package¶
core¶
create_full_problem_to_methods(problems_ids, …) |
Description: Associate all given problems to all given methods. |
run_experiment(problem, method[, metric, …]) |
Description: Initializes the experiment instance. |
run_experiments(problem, method[, metric, …]) |
Description: Initializes the experiment instance. |
metrics¶
mse(y_pred, y_true) |
Description: mean-square-error loss |
cross_entropy(y_pred, y_true[, eps]) |
Description: cross entropy loss, y_pred is equivalent to logits and y_true to labels |
experiment¶
Experiment() |
Description: Streamlines the process of performing experiments and comparing results of methods across a range of problems. |
new_experiment¶
NewExperiment() |
Description: class for implementing algorithms with enforced modularity |
precomputed¶
recompute([verbose, load_bar]) |
Description: Recomputes all the results. |
load_prob_method_to_result([problem_ids, …]) |
Description: Initializes the experiment instance. |
License¶
Some license