Merge pull request #288 from alan-turing-institute/install

installation

Author: mastoffel

.github/workflows/ci.yaml

@@ -52,7 +52,7 @@ jobs:
         run: |
           poetry remove torch
           poetry source add -p explicit pytorch https://download.pytorch.org/whl/cpu
-          poetry add --source pytorch torch
+          poetry add --source pytorch "torch==2.2.1"
 
       - name: Install dependencies
         run: |

README.md

@@ -13,6 +13,8 @@ The project is in early development.
 
 ## Installation
 
+`AutoEmulate` requires Python `>=3.10` and `<3.13`.
+
 There's lots of development at the moment, so we recommend installing the most current version from GitHub:
 
 ```bash

autoemulate/emulators/conditional_neural_process.py

@@ -221,9 +221,9 @@ def fit(self, X, y):
     def predict(self, X, return_std=False):
         check_is_fitted(self)
         X = check_array(X, dtype=np.float32)
-        X_context = torch.from_numpy(self.X_train_).float().unsqueeze(0)
-        y_context = torch.from_numpy(self.y_train_).float().unsqueeze(0)
-        X_target = torch.from_numpy(X).float().unsqueeze(0)
+        X_context = torch.tensor(self.X_train_, dtype=torch.float32).unsqueeze(0)
+        y_context = torch.tensor(self.y_train_, dtype=torch.float32).unsqueeze(0)
+        X_target = torch.tensor(X, dtype=torch.float32).unsqueeze(0)
 
         with torch.no_grad():
             predictions = self.model_.module_.forward(X_context, y_context, X_target)

docs/getting-started/installation.md

@@ -2,6 +2,10 @@
 
 `AutoEmulate` is a Python package that can be installed in a number of ways. In this section we will describe the main ways to install the package.
 
+## Prerequisites
+
+**Python Version:** `AutoEmulate` requires Python `>=3.10` and `<3.13`.
+
 ## Install from GitHub
 
 This is the easiest way to install `AutoEmulate`.
@@ -14,7 +18,7 @@ pip install git+https://github.com/alan-turing-institute/autoemulate.git
 
 ## Install from PyPI
 
-Once we have a release on PyPI, you can install the package from there:
+To get the latest release from PyPI:
 
 ```bash
 pip install autoemulate

tests/models/test_gptorch.py

@@ -17,35 +17,50 @@ def sample_data_y1d():
     return X, y
 
 
+@pytest.fixture
+def new_data_y1d():
+    X, y = make_regression(n_samples=20, n_features=5, n_targets=1, random_state=1)
+    return X, y
+
+
 @pytest.fixture
 def sample_data_y2d():
     X, y = make_regression(n_samples=20, n_features=5, n_targets=2, random_state=0)
     return X, y
 
 
+@pytest.fixture
+def new_data_y2d():
+    X, y = make_regression(n_samples=20, n_features=5, n_targets=2, random_state=1)
+    return X, y
+
+
 # test multitask GP
-def test_multi_output_gpmt(sample_data_y2d):
+def test_multi_output_gpmt(sample_data_y2d, new_data_y2d):
     X, y = sample_data_y2d
     gp = GaussianProcessMT(random_state=42)
     gp.fit(X, y)
-    assert gp.predict(X).shape == (20, 2)
+    X2, _ = new_data_y2d
+    assert gp.predict(X2).shape == (20, 2)
 
 
-def test_predict_with_uncertainty_gpmt(sample_data_y1d):
+def test_predict_with_uncertainty_gpmt(sample_data_y1d, new_data_y1d):
     X, y = sample_data_y1d
     y_shape = y.shape
     gp = GaussianProcessMT(random_state=42)
     gp.fit(X, y)
-    y_pred, y_std = gp.predict(X, return_std=True)
+    X2, _ = new_data_y1d
+    y_pred, y_std = gp.predict(X2, return_std=True)
     assert y_pred.shape == y_shape
     assert y_std.shape == y_shape
 
 
-def test_multitask_gpmt(sample_data_y2d):
+def test_multitask_gpmt(sample_data_y2d, new_data_y2d):
     X, y = sample_data_y2d
     gp = GaussianProcessMT(random_state=42)
     gp.fit(X, y)
-    y_pred, y_std = gp.predict(X, return_std=True)
+    X2, _ = new_data_y2d
+    y_pred, y_std = gp.predict(X2, return_std=True)
     assert y_pred.shape == y.shape
     assert y_std.shape == y.shape
 
@@ -58,34 +73,38 @@ def test_gpmt_param_search(sample_data_y1d):
 
 
 # test multioutput GP
-def test_multioutput_gp(sample_data_y2d):
+def test_multioutput_gp(sample_data_y2d, new_data_y2d):
     X, y = sample_data_y2d
+    X2, _ = new_data_y2d
     gp = GaussianProcess(random_state=42)
     gp.fit(X, y)
     assert gp.predict(X).shape == (20, 2)
 
 
-def test_predict_with_uncertainty_gp(sample_data_y1d):
+def test_predict_with_uncertainty_gp(sample_data_y1d, new_data_y1d):
     X, y = sample_data_y1d
     y_shape = y.shape
     gp = GaussianProcess(random_state=42)
     gp.fit(X, y)
-    y_pred, y_std = gp.predict(X, return_std=True)
+    X2, _ = new_data_y1d
+    y_pred, y_std = gp.predict(X2, return_std=True)
     assert y_pred.shape == y_shape
     assert y_std.shape == y_shape
 
 
-def test_multioutput_gp(sample_data_y2d):
+def test_multioutput_gp(sample_data_y2d, new_data_y2d):
     X, y = sample_data_y2d
+    X2, _ = new_data_y2d
     gp = GaussianProcess(random_state=42)
     gp.fit(X, y)
-    y_pred, y_std = gp.predict(X, return_std=True)
+    y_pred, y_std = gp.predict(X2, return_std=True)
     assert y_pred.shape == y.shape
     assert y_std.shape == y.shape
 
 
-def test_gp_param_search(sample_data_y1d):
+def test_gp_param_search(sample_data_y1d, new_data_y1d):
     X, y = sample_data_y1d
+    X2, _ = new_data_y1d
     em = AutoEmulate()
     em.setup(X, y, models=["gp"], param_search_iters=3)
     em.compare()

tests/test_end_to_end.py

@@ -12,15 +12,15 @@ def kfold():
 
 @pytest.fixture()
 def Xy_single():
-    X = np.random.rand(15, 2)
-    y = np.random.rand(15)
+    X = np.random.rand(30, 2)
+    y = np.random.rand(30)
     return X, y
 
 
 @pytest.fixture()
 def Xy_multi():
-    X = np.random.rand(10, 2)
-    y = np.random.rand(10, 2)
+    X = np.random.rand(30, 2)
+    y = np.random.rand(30, 2)
     return X, y
 
 
@@ -42,7 +42,7 @@ def test_run_param_search(Xy_single, kfold):
         y,
         print_setup=False,
         param_search=True,
-        param_search_iters=2,
+        param_search_iters=1,
         cross_validator=kfold,
     )
     em.compare()

tests/test_estimators.py

@@ -44,5 +44,6 @@
         GaussianProcess(random_state=42),
     ]
 )
+@pytest.mark.filterwarnings("ignore::gpytorch.models.exact_gp.GPInputWarning")
 def test_check_estimator(estimator, check):
     check(estimator)

tests/test_sensitivity_analysis.py

@@ -110,6 +110,7 @@ def test_get_output_names_invalid():
 
 
 # test Sobol analysis ------------------------------------------------------------
+@pytest.mark.filterwarnings("ignore::FutureWarning")
 def test_sobol_analysis(model_1d):
     problem = {
         "num_vars": 2,
@@ -125,6 +126,7 @@ def test_sobol_analysis(model_1d):
     )
 
 
+@pytest.mark.filterwarnings("ignore::FutureWarning")
 def test_sobol_analysis_2d(model_2d):
     problem = {
         "num_vars": 2,
@@ -146,7 +148,8 @@ def sobol_results_1d(model_1d):
     return _sobol_analysis(model_1d, problem)
 
 
-# test conversion to DataFrame --------------------------------------------------
+# # test conversion to DataFrame --------------------------------------------------
+@pytest.mark.filterwarnings("ignore::FutureWarning")
 def test_sobol_results_to_df(sobol_results_1d):
     df = _sobol_results_to_df(sobol_results_1d)
     assert isinstance(df, pd.DataFrame)
@@ -166,11 +169,13 @@ def test_sobol_results_to_df(sobol_results_1d):
 
 
 # test _validate_input ----------------------------------------------------------
+@pytest.mark.filterwarnings("ignore::FutureWarning")
 def test_validate_input(sobol_results_1d):
     with pytest.raises(ValueError):
         _validate_input(sobol_results_1d, "S3")
 
 
+@pytest.mark.filterwarnings("ignore::FutureWarning")
 def test_validate_input_valid(sobol_results_1d):
     Si = _validate_input(sobol_results_1d, "S1")
     assert isinstance(Si, pd.DataFrame)

tests/test_ui.py

@@ -16,8 +16,10 @@
 
 @pytest.fixture
 def param_search_ae():
-    X = np.random.rand(50, 2)
-    y = np.random.rand(50, 1)
+    X = np.random.rand(140, 2)
+    y = np.random.rand(
+        140,
+    )
 
     # names of all models
     all_models = list(model_registry.get_model_names().keys())
@@ -29,7 +31,7 @@ def param_search_ae():
         cross_validator=KFold(n_splits=2),
         param_search_type="random",
         param_search=True,
-        param_search_iters=2,
+        param_search_iters=1,
         models=all_models,
     )
     ae.compare()
@@ -42,7 +44,9 @@ def param_search_ae():
 
 def test_scalers():
     X = np.random.rand(100, 5)
-    y = np.random.rand(100, 1)
+    y = np.random.rand(
+        100,
+    )
 
     scalers = [MinMaxScaler(), RobustScaler()]
 
@@ -56,7 +60,9 @@ def test_scalers():
 
 def test_dimension_reducers():
     X = np.random.rand(100, 10)
-    y = np.random.rand(100, 1)
+    y = np.random.rand(
+        100,
+    )
 
     dim_reducers = [PCA(n_components=5), KernelPCA(n_components=5)]
 
@@ -70,16 +76,15 @@ def test_dimension_reducers():
 
 def test_cross_validators():
     X = np.random.rand(100, 5)
-    y = np.random.rand(100, 1)
-
-    cross_validators = [KFold(n_splits=5)]
+    y = np.random.rand(
+        100,
+    )
 
-    for cross_validator in cross_validators:
-        ae = AutoEmulate()
-        ae.setup(X, y, cross_validator=cross_validator, models=model_subset)
-        ae.compare()
+    ae = AutoEmulate()
+    ae.setup(X, y, cross_validator=KFold(n_splits=5), models=model_subset)
+    ae.compare()
 
-        assert ae.best_model is not None
+    assert ae.best_model is not None
 
 
 def test_param_search(param_search_ae):