Feat/output cost per day in simulator (#88)

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "FSRS-Optimizer"
-version = "4.25.3"
+version = "4.26.0"
 readme = "README.md"
 dependencies = [
     "matplotlib>=3.7.0",

src/fsrs_optimizer/fsrs_optimizer.py

@@ -1283,9 +1283,13 @@ def find_optimal_retention(
         if not verbose:
             return ()
 
-        (_, review_cnt_per_day, learn_cnt_per_day, memorized_cnt_per_day) = simulate(
-            **simulate_config
-        )
+        (
+            _,
+            review_cnt_per_day,
+            learn_cnt_per_day,
+            memorized_cnt_per_day,
+            cost_per_day,
+        ) = simulate(**simulate_config)
 
         def moving_average(data, window_size=365 // 20):
             weights = np.ones(window_size) / window_size
@@ -1324,9 +1328,16 @@ def moving_average(data, window_size=365 // 20):
         ax.legend()
         ax.grid(True)
 
-        fig5 = workload_graph(simulate_config)
+        fig5 = plt.figure()
+        ax = fig5.gca()
+        ax.plot(cost_per_day, label=f"R={self.optimal_retention*100:.0f}%")
+        ax.set_title("Cost per Day")
+        ax.legend()
+        ax.grid(True)
+
+        fig6 = workload_graph(simulate_config)
 
-        return (fig1, fig2, fig3, fig4, fig5)
+        return (fig1, fig2, fig3, fig4, fig5, fig6)
 
     def evaluate(self, save_to_file=True):
         my_collection = Collection(DEFAULT_WEIGHT)
@@ -1870,3 +1881,22 @@ def rmse_matrix(df):
     return mean_squared_error(
         tmp["y"], tmp["p"], sample_weight=tmp["card_id"], squared=False
     )
+
+
+if __name__ == "__main__":
+    model = FSRS(DEFAULT_WEIGHT)
+    stability = torch.tensor([5.0] * 4)
+    difficulty = torch.tensor([1.0, 2.0, 3.0, 4.0])
+    retention = torch.tensor([0.9, 0.8, 0.7, 0.6])
+    rating = torch.tensor([1, 2, 3, 4])
+    state = torch.stack([stability, difficulty]).unsqueeze(0)
+    s_recall = model.stability_after_success(state, retention, rating)
+    print(s_recall)
+    s_forget = model.stability_after_failure(state, retention)
+    print(s_forget)
+
+    retentions = torch.tensor([0.1, 0.2, 0.3, 0.4])
+    labels = torch.tensor([0.0, 1.0, 0.0, 1.0])
+    loss_fn = nn.BCELoss()
+    loss = loss_fn(retentions, labels)
+    print(loss)

src/fsrs_optimizer/fsrs_simulator.py

@@ -61,6 +61,7 @@ def simulate(
     review_cnt_per_day = np.zeros(learn_span)
     learn_cnt_per_day = np.zeros(learn_span)
     memorized_cnt_per_day = np.zeros(learn_span)
+    cost_per_day = np.zeros(learn_span)
 
     def stability_after_success(s, r, d, response):
         hard_penalty = np.where(response == 2, w[15], 1)
@@ -176,7 +177,14 @@ def stability_after_failure(s, r, d):
         review_cnt_per_day[today] = np.sum(true_review)
         learn_cnt_per_day[today] = np.sum(true_learn)
         memorized_cnt_per_day[today] = card_table[col["retrievability"]].sum()
-    return card_table, review_cnt_per_day, learn_cnt_per_day, memorized_cnt_per_day
+        cost_per_day[today] = card_table[col["cost"]][true_review | true_learn].sum()
+    return (
+        card_table,
+        review_cnt_per_day,
+        learn_cnt_per_day,
+        memorized_cnt_per_day,
+        cost_per_day,
+    )
 
 
 def optimal_retention(**kwargs):
@@ -200,7 +208,7 @@ def sample(
 ):
     memorization = []
     for i in range(SAMPLE_SIZE):
-        _, _, _, memorized_cnt_per_day = simulate(
+        _, _, _, memorized_cnt_per_day, _ = simulate(
             w,
             request_retention=r,
             deck_size=deck_size,
@@ -613,7 +621,7 @@ def workload_graph(default_params):
         "first_rating_prob": np.array([0.15, 0.2, 0.6, 0.05]),
         "review_rating_prob": np.array([0.3, 0.6, 0.1]),
     }
-    (_, review_cnt_per_day, learn_cnt_per_day, memorized_cnt_per_day) = simulate(
+    (_, review_cnt_per_day, learn_cnt_per_day, memorized_cnt_per_day, _) = simulate(
         w=default_params["w"],
         max_cost_perday=math.inf,
         learn_limit_perday=10,