Merge branch 'Novartis:main' into main

README.md

@@ -193,11 +193,16 @@ The table below compares the functionalities of `TorchSurv` with those of
 [deepsurv](https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-018-0482-1).
 While several libraries offer survival modelling functionalities, no existing library provides the flexibility to use a custom PyTorch-based neural networks to define the survival model parameters.
 
-The outputs of both the log-likelihood functions and the evaluation metrics functions have undergone thorough comparison with benchmarks generated using Python packages and R packages. The comparisons are summarised in the [Related packages summary](https://opensource.nibr.com/torchsurv/benchmarks.html).
+The outputs of both the log-likelihood functions and the evaluation metrics functions have **undergone thorough comparison with benchmarks generated** using `Python` and `R` packages. The comparisons (at time of publication) are summarised in the [Related packages summary](https://opensource.nibr.com/torchsurv/benchmarks.html).
 
 ![Survival analysis libraries in Python](docs/source/table_python_benchmark.png)
 ![Survival analysis libraries in Python](docs/source/table_python_benchmark_legend.png)
 
+Survival analysis libraries in R. For obtaining the evaluation metrics, packages `survival`, `riskRegression`, `SurvMetrics` and `pec` require the fitted model object as input (a specific object format) and `RisksetROC` imposes a smoothing method. Packages `timeROC`, `riskRegression` and pec force the user to choose a form for subject-specific
+weights (e.g., inverse probability of censoring weighting (IPCW)). Packages `survcomp` and `SurvivalROC` do not implement the general AUC but the censoring-adjusted AUC estimator proposed by Heagerty et al. (2000).
+
+![Survival analysis libraries in R](docs/source/table_r_benchmark.png)
+
 ## Contributing
 
 We value contributions from the community to enhance and improve this project. If you'd like to contribute, please consider the following:
@@ -231,4 +236,4 @@ If you use this project in academic work or publications, we appreciate citing i
       primaryClass={cs.LG},
       doi={https://doi.org/10.48550/arXiv.2404.10761}
 }
-```
+```

dev/environment.yml

@@ -24,3 +24,5 @@ dependencies:
   - nbsphinx=0.9.3
   - ipython=8.20.0
   - twine=4.0.2
+  - scipy=1.12.0
+  - sphinx-math-dollar=1.2.1

docs/AUTHORS.md

@@ -8,4 +8,3 @@ Contributors
 * David Ohlssen <[email protected]> `(contributor)`
 * Berkman Sahiner <[email protected]> `(contributor)`
 * Nicholas Petrick <[email protected]> `(contributor)`
-

docs/conf.py

@@ -28,8 +28,10 @@
     "nbsphinx",
     "sphinx.ext.viewcode",
     "sphinxcontrib.bibtex",
+    "sphinx_math_dollar",
 ]
 
+
 # templates_path = ['_templates']
 exclude_patterns = ["_build", "Thumbs.db", ".DS_Store"]
 

docs/index.rst

@@ -16,6 +16,7 @@ Welcome to TorchSurv's documentation!
     :maxdepth: 2
     :caption: Tutorials:
 
+    notebooks/survival
     notebooks/introduction
     notebooks/momentum
 

src/torchsurv/metrics/auc.py

@@ -1192,7 +1192,6 @@ def _find_torch_unique_indices(
     def _validate_auc_inputs(
         estimate, time, auc_type, new_time, weight, weight_new_time
     ):
-
         # check new_time and weight are provided, weight_new_time should be provided
         if all([new_time is not None, weight is not None, weight_new_time is None]):
             raise ValueError(
@@ -1222,12 +1221,10 @@ def _update_auc_new_time(
         weight: torch.tensor,
         weight_new_time: torch.tensor,
     ) -> torch.tensor:
-
         # update new time
         if (
             new_time is not None
         ):  # if new_time are specified: ensure it has the correct format
-
             # ensure that new_time are float
             if isinstance(new_time, int):
                 new_time = torch.tensor([new_time]).float()
@@ -1237,7 +1234,6 @@ def _update_auc_new_time(
                 new_time = new_time.unsqueeze(0)
 
         else:  # else: find new_time
-
             # if new_time are not specified, use unique event time
             mask = event & (time < torch.max(time))
             new_time, inverse_indices, counts = torch.unique(
@@ -1261,7 +1257,6 @@ def _update_auc_new_time(
     def _update_auc_estimate(
         estimate: torch.tensor, new_time: torch.tensor
     ) -> torch.tensor:
-
         # squeeze estimate if shape = (n_samples, 1)
         if estimate.ndim == 2 and estimate.shape[1] == 1:
             estimate = estimate.squeeze(1)
@@ -1281,7 +1276,6 @@ def _update_auc_weight(
         weight: torch.tensor,
         weight_new_time: torch.tensor,
     ) -> torch.tensor:
-
         # if weight was not specified, weight of 1
         if weight is None:
             weight = torch.ones_like(time)

src/torchsurv/metrics/brier_score.py

@@ -171,10 +171,13 @@ def __call__(
         )
 
         # update inputs as required
-        estimate, new_time, weight, weight_new_time = (
-            BrierScore._update_brier_score_new_time(
-                estimate, time, new_time, weight, weight_new_time
-            )
+        (
+            estimate,
+            new_time,
+            weight,
+            weight_new_time,
+        ) = BrierScore._update_brier_score_new_time(
+            estimate, time, new_time, weight, weight_new_time
         )
         weight, weight_new_time = BrierScore._update_brier_score_weight(
             time, new_time, weight, weight_new_time
@@ -190,14 +193,14 @@ def __call__(
 
         # Calculating the residuals for each subject and time point
         residuals = torch.zeros_like(estimate)
-        for i, t in enumerate(new_time):
-            est = estimate[:, i]
-            is_case = ((time <= t) & (event)).int()
-            is_control = (time > t).int()
+        for index, new_time_i in enumerate(new_time):
+            est = estimate[:, index]
+            is_case = ((time <= new_time_i) & (event)).int()
+            is_control = (time > new_time_i).int()
 
-            residuals[:, i] = (
+            residuals[:, index] = (
                 torch.square(est) * is_case * weight
-                + torch.square(1.0 - est) * is_control * weight_new_time[i]
+                + torch.square(1.0 - est) * is_control * weight_new_time[index]
             )
 
         # Calculating the brier scores at each time point
@@ -827,7 +830,6 @@ def _validate_brier_score_inputs(
         weight: torch.tensor,
         weight_new_time: torch.tensor,
     ) -> torch.tensor:
-
         # check new_time and weight are provided, weight_new_time should be provided
         if all([new_time is not None, weight is not None, weight_new_time is None]):
             raise ValueError(
@@ -859,7 +861,6 @@ def _update_brier_score_new_time(
         weight: torch.tensor,
         weight_new_time: torch.tensor,
     ) -> torch.tensor:
-
         # check format of new_time
         if (
             new_time is not None
@@ -871,7 +872,6 @@ def _update_brier_score_new_time(
                 new_time = new_time.unsqueeze(0)
 
         else:  # else: find new_time
-
             # if new_time are not specified, use unique time
             new_time, inverse_indices, counts = torch.unique(
                 time, sorted=True, return_inverse=True, return_counts=True
@@ -896,7 +896,6 @@ def _update_brier_score_weight(
         weight: torch.tensor,
         weight_new_time: torch.tensor,
     ) -> torch.tensor:
-
         # if weight was not specified, weight of 1
         if weight is None:
             weight = torch.ones_like(time)

src/torchsurv/metrics/cindex.py

@@ -1,5 +1,6 @@
 import copy
 import sys
+import warnings
 from typing import Optional, Tuple
 
 import torch
@@ -639,6 +640,13 @@ def _compare_noether(self, other):
         cindex1_se = self._concordance_index_se()
         cindex2_se = other._concordance_index_se()
 
+        # Suppress the specific warning
+        warnings.filterwarnings(
+            "ignore",
+            message="Metric `SpearmanCorrcoef` will save all targets and predictions in the buffer. For large datasets, this may lead to large memory footprint.",
+            category=UserWarning,
+        )
+
         # compute spearman correlation between risk prediction
         corr = regression.SpearmanCorrCoef()(
             self.estimate.reshape(-1), other.estimate.reshape(-1)

tests/test_cox.py

@@ -57,9 +57,7 @@ def test_boolean_y(self):
     def test_log_likelihood_without_ties(self):
         """test cox partial log likelihood without ties on lung and gbsg datasets"""
         for benchmark_cox_loglik in benchmark_cox_logliks:
-
             if benchmark_cox_loglik["no_ties"][0] == True:
-
                 log_lik = -cox(
                     torch.tensor(
                         benchmark_cox_loglik["log_hazard"], dtype=torch.float32
@@ -82,9 +80,7 @@ def test_log_likelihood_without_ties(self):
     def test_log_likelihood_with_ties(self):
         """test Efron and Breslow's approximation of cox partial log likelihood with ties on lung and gbsg data"""
         for benchmark_cox_loglik in benchmark_cox_logliks:
-
             if benchmark_cox_loglik["no_ties"][0] == False:
-
                 # efron approximation of partial log likelihood
                 log_lik_efron = -cox(
                     torch.tensor(

tests/test_kaplan_meier.py

@@ -27,7 +27,6 @@ def test_kaplan_meier_survival_distribution_real_data(self):
         """test Kaplan Meier survival distribution estimate on lung and gbsg datasets"""
 
         for benchmark_kaplan_meier in benchmark_kaplan_meiers:
-
             event = torch.tensor(benchmark_kaplan_meier["status"]).bool()
             time = torch.tensor(benchmark_kaplan_meier["time"], dtype=torch.float32)
             new_time = torch.tensor(
@@ -209,9 +208,9 @@ def test_kaplan_meier_prediction_error_raised(self):
         for batch in batch_container.batches:
             (train_time, train_event, test_time, *_) = batch
 
-            train_event[-1] = (
-                False  # if last event is censoring, the last KM is > 0 and it cannot predict beyond this time
-            )
+            train_event[
+                -1
+            ] = False  # if last event is censoring, the last KM is > 0 and it cannot predict beyond this time
             km = KaplanMeierEstimator()
             km(train_event, train_time, censoring_dist=False)
 

tests/test_momentum.py

@@ -13,7 +13,6 @@
 
 
 class TestMometum(unittest.TestCase):
-
     def test_momentum_weibull(self):
         model = Momentum(
             backbone=nn.Sequential(nn.Linear(8, 2)),  # Weibull expect two ouputs

tests/utils.py

@@ -279,7 +279,6 @@ def get_input_array(self) -> Tuple[np.array, np.array, np.array, np.array]:
         )
 
     def _generate_input(self):
-
         # random maximum time in observational period
         tmax = torch.randint(5, 500, (1,)).item()
 
@@ -297,7 +296,6 @@ def _generate_input(self):
         self._generate_new_time()
 
     def _generate_data(self, tmax: int, n_train: int, n_test: int):
-
         # time-to-event or censoring in train
         train_time = torch.randint(1, tmax + 1, (n_train,)).float()
 
@@ -340,7 +338,6 @@ def _generate_data(self, tmax: int, n_train: int, n_test: int):
     def _enforce_conditions_data(
         self, time: torch.tensor, event: torch.tensor, dataset_type: str
     ) -> Tuple[torch.tensor, torch.tensor]:
-
         # if test max time should be greater than train max time
         if dataset_type == "test":
             if self.test_max_time_gt_train_max_time:
@@ -395,15 +392,13 @@ def _enforce_conditions_data(
         return time, event
 
     def _generate_estimate(self):
-
         # random risk score for observations in test
         estimate = torch.randn(len(self.test_event))
 
         # enforce conditions risk score
         self.estimate = self._enforce_conditions_estimate(estimate)
 
     def _enforce_conditions_estimate(self, estimate: torch.tensor) -> torch.tensor:
-
         # if there should be ties in risk score associated to patients with event
         if self.ties_score_events:
             estimate[torch.where(self.test_event == 1.0)[0][0]] = estimate[
@@ -425,7 +420,6 @@ def _enforce_conditions_estimate(self, estimate: torch.tensor) -> torch.tensor:
         return estimate
 
     def _generate_new_time(self):
-
         if torch.all(self.test_event == False):
             # if all patients are censored in test, no evaluation time
             new_time = torch.tensor([])
@@ -447,7 +441,6 @@ def _generate_new_time(self):
         self.new_time = self._enforce_conditions_time(new_time)
 
     def _enforce_conditions_time(self, new_time: torch.tensor) -> torch.tensor:
-
         # if the test max time should be included in evaluation time
         if self.test_max_time_in_new_time:
             new_time = torch.cat(
@@ -457,7 +450,6 @@ def _enforce_conditions_time(self, new_time: torch.tensor) -> torch.tensor:
         return new_time
 
     def _evaluate_conditions(self):
-
         # are there ties in event times
         self.has_train_ties_time_event = self._has_ties(
             self.train_time[self.train_event == 1]
@@ -614,7 +606,6 @@ def generate_batches(self, n_batch: int, flags_to_set: list):
             n_batch = len(flags_to_set)
 
         for i in range(n_batch):
-
             if i >= len(flags_to_set):
                 # simulate data without flag
                 self.generate_one_batch()