autots 0.0.2__py3-none-any.whl

AutoTS-0.0.2.dist-info/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2019 Colin Catlin
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

AutoTS-0.0.2.dist-info/METADATA

@@ -0,0 +1,176 @@
+Metadata-Version: 2.1
+Name: AutoTS
+Version: 0.0.2
+Summary: Automated Time Series Forecasting
+Home-page: https://github.com/winedarksea/AutoTS
+Author: Colin Catlin
+Author-email: colin.catlin@syllepsis.live
+License: MIT
+Platform: UNKNOWN
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.6
+Description-Content-Type: text/markdown
+Requires-Dist: numpy
+Requires-Dist: pandas
+Requires-Dist: statsmodels
+Requires-Dist: scikit-learn
+Requires-Dist: holidays
+Provides-Extra: additional_models
+Requires-Dist: fbprophet ; extra == 'additional_models'
+Requires-Dist: fredapi ; extra == 'additional_models'
+
+# AutoTS
+Unstable prototype: version 0.0.2
+### Project CATS (Catlin Automated Time Series)
+(or maybe eventually: Clustered Automated Time Series)
+#### Model Selection for Multiple Time Series
+
+Simple package for comparing and predicting with open-source time series implementations.
+For other time series needs, check out the package list here: https://github.com/MaxBenChrist/awesome_time_series_in_python
+
+`pip install autots`
+#### Requirements:
+	Python >= 3.5 (typing) >= 3.6 (GluonTS)
+	pandas
+	sklearn >= 0.20.0 (ColumnTransformer)
+	statsmodels
+	holidays
+
+`pip install autots['additional models']`
+#### Requirements
+	fbprophet
+	fredapi (example datasets)
+
+## Basic Use
+Input data is expected to come in a 'long' format with three columns: Date (ideally already in pd.DateTime format), Value, and Series ID. the column name for each of these is passed to .fit(). For a single time series, series_id can be = None. 
+
+```
+from autots.datasets import load_toy_daily
+df_long = load_toy_daily()
+
+from autots import AutoTS
+model = AutoTS(forecast_length = 14, frequency = 'infer',
+               prediction_interval = 0.9, ensemble = True, weighted = False,
+               max_generations = 5, num_validations = 2, validation_method = 'even')
+model = model.fit(df_long, date_col = 'date', value_col = 'value', id_col = 'series_id' )
+
+# Print the name of the best mode
+print(model.best_model['Model'].iloc[0])
+
+prediction = model.predict()
+# point forecasts dataframe
+forecasts_df = prediction.forecast
+# accuracy of all tried model results (not including cross validation)
+model_results = model.main_results.model_results
+```
+
+## Underlying Process
+AutoTS works in the following way at present:
+* It begins by taking long data and converting it to a wide dataframe with DateTimeIndex
+* An initial train/test split is generated where the test is the most recent data, of forecast_length
+* A random template of models is generated and tested on the initial train/test
+	* Models consist of a pre-transformation step (fill na options, outlier removal options, etc), and algorithm (ie ETS) and model paramters (trend, damped, etc)
+* The top models (selected by a combination of SMAPE, MAE, RMSE) are recombined with random mutations for n_generations
+* A handful of the best models from this process go to cross validation, where they are re-assessed on new train/test splits.
+* The best model in validation is selected as best_model and used in the .predict() method to generate forecasts.
+
+## Caveats and Advice
+
+#### Short Training History
+How much data is 'too little' depends on the seasonality and volatility of the data. 
+But less than half a year of daily data or less than two years of monthly data are both going to be tight. 
+Minimal training data most greatly impacts the ability to do proper cross validation. Set num_validations = 0 in such cases. 
+Since ensembles are based on the test dataset, it would also be wise to set ensemble = False if num_validations = 0.
+
+#### Too much training data.
+Too much data is already handled to some extent by 'context_slicer' in the transformations, which tests using less training data. 
+That said, large datasets will be slower and more memory intensive, for high frequency data (say hourly) it can often be advisable to roll that up to a higher level (daily, hourly, etc.). 
+Rollup can be accomplished by specifying the frequency = your rollup frequency, and then setting the agg_func = 'sum' or 'mean' or other appropriate statistic.
+
+#### Lots of NaN in data
+Various NaN filling techniques are tested in the transformation. Rolling up data to a lower frequency may also help deal with NaNs.
+
+#### More than one preord regressor
+'Preord' regressor stands for 'Preordained' regressor, to make it clear this is data that will be know with high certainy about the future. 
+Such data about the future is rare, one example might be number of stores that will be (planned to be) open each given day in the future when forecast sales. 
+Since many algorithms do not handle more than one regressor, only one is handled here. If you would like to use more than one, 
+manually select the best variable or use dimensionality reduction to reduce the features to one dimension. 
+However, the model can handle quite a lot of parallel time series. Additional regressors can be passed through as additional time series to forecast. 
+The regression models here can utilize the information they provide to help improve forecast quality. 
+To prevent forecast accuracy for considering these additional series too heavily, input series weights that lower or remove their forecast accuracy from consideration.
+
+#### Categorical Data
+Categorical data is handled, but it is handled poorly. For example, optimization metrics do not currently include any categorical accuracy metrics. 
+For categorical data that has a meaningful order (ie 'low', 'medium', 'high') it is best for the user to encode that data before passing it in, 
+thus properly capturing the relative sequence (ie 'low' = 1, 'medium' = 2, 'high' = 3).
+
+#### Custom Metrics
+Implementing new metrics is rather difficult. However the internal 'Score' that compares models can easily be adjusted by passing through custom metric weights. 
+Higher weighting increases the importance of that metric. 
+`metric_weighting = {'smape_weighting' : 9, 'mae_weighting' : 1, 'rmse_weighting' : 5, 'containment_weighting' : 1, 'runtime_weighting' : 0.5}` 
+sMAPE is generally the most versatile across multiple series, but doesn't handle forecasts with lots of zeroes well. 
+Contaiment measures the percent of test data that falls between the upper and lower forecasts. 
+
+## To-Do
+* Smaller
+	* Recombine best two of each model, if two or more present
+	* Duplicates still seem to be occurring in the genetic template runs
+	* Inf appearing in MAE and RMSE (possibly all NaN in test)
+	* Na Tolerance for test in simple_train_test_split
+	* Relative/Absolute Imports and reduce package reloading
+	* User regressor to sklearn model regression_type
+	* Import/export template
+	* ARIMA + Detrend fails
+* Things needing testing:
+    * Confirm per_series weighting works properly
+    * Passing in Start Dates - (Test)
+    * Different frequencies
+    * Various verbose inputs
+    * Test holidays on non-daily data
+    * Handle categorical forecasts where forecast leaves known values
+* Speed improvements, Profiling, Parallelization, and Distributed options for general greater speed
+* Generate list of functional frequences, and improve usability on rarer frequenices
+* Warning/handling if lots of NaN in most recent (test) part of data
+* Figures: Add option to output figures of train/test + forecast, other performance figures
+* Input and Output saved templates as .csv and .json
+* 'Check Package' to check if optional model packages are installed
+* Pre-clustering on many time series
+* If all input are Int, convert floats back to int
+* Trim whitespace on string inputs
+* Hierachial correction (bottom-up to start with)
+* Improved verbosity controls and options. Replace most 'print' with logging.
+* Export as simpler code (as TPOT)
+* AIC metric, other accuracy metrics
+* Analyze and return inaccuracy patterns (most inaccurate periods out, days of week, most inaccurate series)
+* Used saved results to resume a search partway through
+* Generally improved probabilistic forecasting
+* Option to drop series which haven't had a value in last N days
+* Option to change which metric is being used for model selections
+* Use quantile of training data to provide upper/lower forecast for Last Value Naive (so upper forecast might be 95th percentile largest number)
+* More thorough use of setting random seed
+* For monthly data account for number of days in month
+* Option to run generations until generations no longer see improvement of at least X % over n generations
+
+#### New Ensembles:
+	best 3 (unique algorithms not just variations)
+	forecast distance 30/30/30
+	best per series ensemble
+	best point with best probalistic containment
+#### New models:
+	Seasonal Naive
+	Last Value + Drift Naive
+	Simple Decomposition forecasting
+	GluonTS Models
+	Simulations
+	Sklearn + TSFresh
+	Sklearn + polynomial features
+	Sktime
+	Ta-lib
+	tslearn
+	pydlm
+	Isotonic regression
+	TPOT if it adds multioutput functionality
+
+

AutoTS-0.0.2.dist-info/RECORD

@@ -0,0 +1,26 @@
+autots/__init__.py,sha256=YGDDQCYlYZZUc8V3BTB9TFrCKLdPcHrtwIbfNIPksvA,396
+autots/datasets/__init__.py,sha256=mu6O826oBODAPYN-fjAVdWQLPqxcwF4LHvyKyJZZno8,123
+autots/datasets/_base.py,sha256=uIriLBZh8HLxOK55243e1uvcXoQVngNv8UJpTduvHIU,483
+autots/datasets/fred.py,sha256=vtp2MdbybZ8kH5jXApKbmG7N1OnAMBuYvFd-5Eaz3x0,2322
+autots/datasets/data/toy_daily.csv,sha256=WHR6jq7vE35fW6gNyKMDtphMiSBNDfSn0sD3-hu6nkg,13090
+autots/evaluator/__init__.py,sha256=W52tlP7cKjwNBkvV9F9m8pNYk6QmVMvfUz9rOzb7c_w,30
+autots/evaluator/auto_model.py,sha256=uApYbzRLpoQ-74C1Y1X8GRivkeEbwWUmjVvoVGjXdpc,36466
+autots/evaluator/auto_ts.py,sha256=gm1p2_a1Kor9kd1iEVbOzmHLQjakdbqsrLYdfEUUMVQ,31813
+autots/evaluator/metrics.py,sha256=at1eeey9Cik8LvTiv1Mg3bsc3FKH0Sb8S2SwNvJJPq0,6039
+autots/models/basics.py,sha256=3wPcdNHKwSFqjFXKndl8r_8WIjAFDIjk8nIaxt2yL6g,10072
+autots/models/ensemble.py,sha256=DqirKJru0UGDu2XwKM2xeNVelihaGl-Y6HirZmITyN0,9207
+autots/models/prophet.py,sha256=gFUz4I4YA4l19NjFhQGFfeouyabMaJKgCUO8-eFvtCo,6926
+autots/models/sklearn.py,sha256=mzoLn1P9UGY6edHwIAeSJAE53GNOvXQ3gkDvjxbKlTs,8420
+autots/models/statsmodels.py,sha256=vBvzESJf8WUnC_JiP8W4Z3qFLQ9RgrKTj72kVwGTZDA,16585
+autots/tools/__init__.py,sha256=d7csIB5IR_Bty-B5evzIbB3KRtPWMoSY4SPp7SdFerA,27
+autots/tools/holiday.py,sha256=66iSw3MFZSsDiWFxu3pq_-y_TnwY96C2HKWqvQW3wOE,1892
+autots/tools/impute.py,sha256=UNMkw9m9nW5umLXQJgAbwXRyicALNWzqWNt8f_6iabs,3301
+autots/tools/probabilistic.py,sha256=YV_bHSIwoELs6Mqu_WA6E_jAbTuvFfWQPKWuPrImKiY,4490
+autots/tools/profile.py,sha256=WfszidiNKSggk6J7vKKu7Atvfxa7ioi-RN-YWbpA3Do,935
+autots/tools/shaping.py,sha256=5DAmr-D5pCk3Kx0OqOvDglGyxToQYQqPH1Ku4H6dOwU,11660
+autots/tools/transform.py,sha256=qCapdKCVQpAPMvSWj5nyY-aO_D1Ej99WOPb_ywpg7tc,16659
+AutoTS-0.0.2.dist-info/LICENSE,sha256=8UOx7AhGiXXInHrdKEY1IDVHiG3t8Sg817EVjn9MXks,1090
+AutoTS-0.0.2.dist-info/METADATA,sha256=UKiBBQqmFIv1n6nJucFIdMtZ6m4tOoh-rP5-0wZJqoU,8977
+AutoTS-0.0.2.dist-info/WHEEL,sha256=v8slff5hmCpvciQ3G55d2d1CnOBupjDFJHDE2dUb1Ao,97
+AutoTS-0.0.2.dist-info/top_level.txt,sha256=bhy_5vpZM1lRYmoAfSD1mnUSNOyfnKBoihQmH6pEzAI,7
+AutoTS-0.0.2.dist-info/RECORD,,

AutoTS-0.0.2.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: bdist_wheel (0.33.6)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

AutoTS-0.0.2.dist-info/top_level.txt

@@ -0,0 +1 @@
+autots

autots/__init__.py

@@ -0,0 +1,16 @@
+"""
+Automated Time Series Model Selection for Python
+
+https://github.com/winedarksea/AutoTS
+"""
+from autots.datasets import load_toy_daily
+from autots.evaluator.auto_ts import AutoTS
+
+__version__ = '0.0.2'
+
+__all__ = ['load_toy_daily', 'AutoTS']
+
+# import logging
+# logger = logging.getLogger(__name__)
+# logger.addHandler(logging.StreamHandler())
+# logger.setLevel(logging.INFO)

autots/datasets/__init__.py

@@ -0,0 +1,6 @@
+"""
+Tools for Importing Sample Data
+"""
+from autots.datasets._base import load_toy_daily
+
+__all__ = ['load_toy_daily']

autots/datasets/_base.py

@@ -0,0 +1,16 @@
+from os.path import dirname, join
+import numpy as np
+import pandas as pd
+
+def load_toy_daily():
+    """
+    4 series of sample daily data from late 2019
+    Testing some basic missing and categorical features.
+    """
+    module_path = dirname(__file__)
+    data_file_name = join(module_path, 'data', 'toy_daily.csv')
+    
+    df_long = pd.read_csv(data_file_name)
+    df_long['date'] = pd.to_datetime(df_long['date'], infer_datetime_format = True)
+
+    return df_long