Switch from using prophet-prebuilt to prophet (#4045)

* Switch prophet-prebuilt to prophet in pyproject.toml

* (Test) add git-test-prophet to windows tests

* Actually install prophet for the windows prophet tests

* Fix typo

* Remove now unnecessary install comments

* Test updates to get windows prophet tests passing

* Hopefully final fix for windows prophet tests

* First attempt at adding prophet to conda build

* Add prophet to regular pyproject dependencies (prophet extra exists, redundant)

* GH workflows should no longer need to manuall install prophet

* Dependency checks need to track prophet now too

* Dont pin prophet in pyproject.toml

* Release notes

* Theres a newer version of prophet

* Back out conda changes

* Add back prophet as a rtd extra

* Remove prophet from windows automl GH workflows to decrease runtime

* Temporarily pin woodwork for testing purposes

* Move non-prophet-dependent tests from git-test-prophet to git-test-other

* Add a few files back for consistency

* Unpin woodwork

* Update release_notes.rst

---------

Co-authored-by: chukarsten <64713315+chukarsten@users.noreply.github.com>

Author: eccabay

.github/workflows/windows_nightlies.yaml

@@ -13,7 +13,7 @@ jobs:
       fail-fast: false
       matrix:
         python_version: ['3.8', '3.9']
-        command: ['git-test-automl', 'git-test-modelunderstanding', 'git-test-other', 'git-test-parallel']
+        command: ['git-test-automl', 'git-test-modelunderstanding', 'git-test-other', 'git-test-parallel', 'git-test-prophet']
     steps:
       - name: Download Miniconda
         shell: pwsh
@@ -57,7 +57,18 @@ jobs:
           conda config --add channels conda-forge
           conda activate curr_py
           conda install numba -q -y
-      - name: Install EvalML with test requirements
+      - if: ${{ matrix.command == 'git-test-prophet' }}
+        name: Install EvalML with test requirements and prophet
+        shell: pwsh
+        run: |
+          . $env:USERPROFILE\Miniconda3\shell\condabin\conda-hook.ps1
+          conda activate curr_py
+          python -m pip install --upgrade pip
+          python -m pip install .[test]
+          python -m pip install .[prophet]
+          pip freeze
+      - if: ${{ matrix.command != 'git-test-prophet' }}
+        name: Install EvalML with test requirements
         shell: pwsh
         run: |
           . $env:USERPROFILE\Miniconda3\shell\condabin\conda-hook.ps1

.github/workflows/windows_unit_tests.yaml

@@ -15,7 +15,7 @@ jobs:
       fail-fast: false
       matrix:
         python_version: ['3.8']
-        command: ['git-test-automl', 'git-test-modelunderstanding', 'git-test-other', 'git-test-parallel']
+        command: ['git-test-automl', 'git-test-modelunderstanding', 'git-test-other', 'git-test-parallel', 'git-test-prophet']
     steps:
       - name: Download Miniconda
         shell: pwsh
@@ -59,7 +59,18 @@ jobs:
           conda config --add channels conda-forge
           conda activate curr_py
           conda install numba -q -y
-      - name: Install EvalML with test requirements
+      - if: ${{ matrix.command == 'git-test-prophet' }}
+        name: Install EvalML with test requirements and prophet
+        shell: pwsh
+        run: |
+          . $env:USERPROFILE\Miniconda3\shell\condabin\conda-hook.ps1
+          conda activate curr_py
+          python -m pip install --upgrade pip
+          python -m pip install .[test]
+          python -m pip install .[prophet]
+          pip freeze
+      - if: ${{ matrix.command != 'git-test-prophet' }}
+        name: Install EvalML with test requirements
         shell: pwsh
         run: |
           . $env:USERPROFILE\Miniconda3\shell\condabin\conda-hook.ps1

Makefile

@@ -52,12 +52,12 @@ git-test-modelunderstanding:
 
 .PHONY: git-test-other
 git-test-other:
-	pytest evalml/tests --ignore evalml/tests/automl_tests/ --ignore evalml/tests/tuner_tests/ --ignore evalml/tests/model_understanding_tests/ --ignore evalml/tests/pipeline_tests/ --ignore evalml/tests/utils_tests/ --ignore evalml/tests/component_tests/test_prophet_regressor.py --ignore evalml/tests/component_tests/test_components.py --ignore evalml/tests/component_tests/test_utils.py --ignore evalml/tests/integration_tests/ -n 2 --durations 0 --timeout $(TIMEOUT) --cov=evalml --cov-config=./pyproject.toml --junitxml=test-reports/git-test-other-junit.xml
+	pytest evalml/tests --ignore evalml/tests/automl_tests/ --ignore evalml/tests/tuner_tests/ --ignore evalml/tests/model_understanding_tests/ --ignore evalml/tests/pipeline_tests/test_pipelines.py --ignore evalml/tests/component_tests/test_prophet_regressor.py --ignore evalml/tests/component_tests/test_components.py --ignore evalml/tests/component_tests/test_utils.py --ignore evalml/tests/integration_tests/ -n 2 --durations 0 --timeout $(TIMEOUT) --cov=evalml --cov-config=./pyproject.toml --junitxml=test-reports/git-test-other-junit.xml
 	make doctests
 
 .PHONY: git-test-prophet
 git-test-prophet:
-	pytest evalml/tests/component_tests/test_prophet_regressor.py evalml/tests/component_tests/test_components.py evalml/tests/component_tests/test_utils.py evalml/tests/pipeline_tests/ evalml/tests/utils_tests/ -n 2 --durations 0 --timeout $(TIMEOUT) --cov=evalml --cov-config=./pyproject.toml --junitxml=test-reports/git-test-prophet-junit.xml
+	pytest evalml/tests/component_tests/test_prophet_regressor.py evalml/tests/component_tests/test_components.py evalml/tests/component_tests/test_utils.py evalml/tests/pipeline_tests/test_pipelines.py -n 2 --durations 0 --timeout $(TIMEOUT) --cov=evalml --cov-config=./pyproject.toml --junitxml=test-reports/git-test-prophet-junit.xml
 
 .PHONY: git-test-integration
 git-test-integration:

docs/source/install.md

@@ -96,35 +96,11 @@ $ conda install -c conda-forge alteryx-open-src-update-checker
 ## Time Series support with Facebook's Prophet
 
 To support the `Prophet` time series estimator, be sure to install it as an extra requirement. Please note that this may take a few minutes.
-Prophet is currently only supported via pip installation in EvalML for Mac with CmdStan as a backend.
 ```shell
 pip install evalml[prophet]
 ```
 Another option for installing Prophet with CmdStan as a backend is to use `make installdeps-prophet`.
 
-Note: In order to do this, you must have the EvalML repo cloned and you must be in the top level folder `<your_directory>/evalml/` to execute this command.
-This command will do the following:
-- Pip install `cmdstanpy==0.9.68`
-- Execute the `install_cmdstan.py` script found within your `site-packages/cmdstanpy` which builds `cmdstan` in your `site-packages`.
-- Install `Prophet==1.0.1` with the `CMDSTAN` and `STAN_BACKEND` environment variables set.
-
-If the `site-packages` path is incorrect or you'd like to specify a different one, just run `make installdeps-prophet SITE_PACKAGES_DIR="<path_to_your_site_packages>"`.
-
-If you'd like to have more fine-tuned control over the installation steps for Prophet, such as specifying the backend, follow these steps:
-
-````{tab} PyStan (default)
-```console
-$ pip install prophet==1.0.1
-```
-````
-````{tab} CmdStanPy backend
-```console
-$ pip install cmdstanpy==0.9.68
-$ python <path_to_installed_cmdstanpy>/install_cmdstan.py --dir <path_to_build_cmdstan> -v <version_to_use>
-$ CMDSTAN=<path_to_build_cmdstan>/cmdstan-<version_to_use> STAN_BACKEND=CMDSTANPY pip install prophet==1.0.1
-
-```
-````
 
 ## Windows Additional Requirements & Troubleshooting
 

docs/source/release_notes.rst

@@ -6,6 +6,7 @@ Release Notes
     * Changes
         * Added Oversampler nullable type incompatibility in X :pr:`4068`
         * Removed nullable handling from objective functions, ``roc_curve``, and ``correlation_matrix`` :pr:`4072`
+        * Transitioned from ``prophet-prebuilt`` to ``prophet`` directly :pr:`4045`
     * Documentation Changes
     * Testing Changes
 

evalml/pipelines/components/estimators/regressors/xgboost_regressor.py

@@ -117,7 +117,7 @@ def get_prediction_intervals(
         coverage: List[float] = None,
         predictions: pd.Series = None,
     ) -> Dict[str, pd.Series]:
-        """Find the prediction intervals using the fitted ProphetRegressor.
+        """Find the prediction intervals using the fitted XGBoostRegressor.
 
         Args:
             X (pd.DataFrame): Data of shape [n_samples, n_features].

evalml/tests/automl_tests/test_automl.py

@@ -4935,7 +4935,6 @@ def test_automl_restricts_use_covariates_for_arima(
     hyperparams,
     automl_algo,
     AutoMLTestEnv,
-    is_using_windows,
 ):
     X, y = datasets.make_classification(
         n_samples=n_rows,
@@ -4993,7 +4992,6 @@ def test_automl_does_not_restrict_use_covariates_if_user_specified(
     hyperparams,
     automl_algo,
     AutoMLTestEnv,
-    is_using_windows,
     X_y_binary,
 ):
     X, y = X_y_binary

evalml/tests/component_tests/test_utils.py

@@ -118,7 +118,7 @@ def test_all_components(
     is_using_conda,
 ):
     if is_using_conda:
-        # No prophet, ARIMA, and vowpalwabbit
+        # No prophet
         expected_components = all_requirements_set.difference(
             not_supported_in_conda,
         )

evalml/tests/pipeline_tests/test_pipeline_utils.py

@@ -856,7 +856,12 @@ def test_generate_pipeline_example(
     X_y_multi,
     X_y_regression,
     ts_data,
+    is_using_windows,
 ):
+    if is_using_windows:
+        pytest.xfail(
+            "Pipeline code generation is not expected to work on Windows machines",
+        )
     if automl_type == ProblemTypes.BINARY:
         X, y = X_y_binary
     elif automl_type == ProblemTypes.MULTICLASS:

evalml/tests/pipeline_tests/test_pipelines.py

@@ -111,15 +111,11 @@ def test_allowed_model_families():
 
 def test_all_estimators(
     is_using_conda,
-    is_using_windows,
 ):
     if is_using_conda:
         n_estimators = 17
     else:
-        # This is wrong because only prophet is missing in windows
-        # but we don't run this test in windows.
-        # TODO: Change when https://github.com/alteryx/evalml/issues/3190 is addressed
-        n_estimators = 16 if is_using_windows else 18
+        n_estimators = 18
     assert len(_all_estimators_used_in_search()) == n_estimators