began argmin_days_since

Author: quinn-dougherty

src/penn_chime/charts.py

@@ -109,6 +109,9 @@ def additional_projections_chart(
     })
     dat["day"] = dat.index
 
+    if parameters.n_days_since_first_hospitalized:
+        dat["day"] = dat.day - parameters.n_days_since_first_hospitalized
+
     as_date = parameters.as_date
     max_y_axis = parameters.max_y_axis
 

src/penn_chime/models.py

@@ -8,6 +8,7 @@
 from __future__ import annotations
 
 from typing import Dict, Generator, Tuple, Optional
+from datetime import date, timedelta
 
 import numpy as np  # type: ignore
 import pandas as pd  # type: ignore
@@ -98,16 +99,16 @@ def __init__(self, p: Parameters):
         self.census_df = census_df
 
         #if p.n_days_since_first_hospitalized is None:
-        #    continue
-
+        #    p.n_days_since_first_hospitalized = int(self.argmin_ds(p))
+        #    p.date_first_hospitalized = date.today() - timedelta(days=p.n_days_since_first_hospitalized)
         if p.n_days_since_first_hospitalized is not None and p.doubling_time is None:
             # optimize doubling_time
             argmin_dt = None
             min_loss = 2.0**99
             censes = dict()
             current_infecteds = dict()
             for dt in np.linspace(1,15,29):
-                censes[dt], current_infecteds[dt] = self.run_projection(p, dt)
+                censes[dt], current_infecteds[dt] = self.run_projection_dt(p, dt)
                 # self.current_infected = current_infecteds[dt] # log it into state for no reason really
                 self.census_df = censes[dt] # log it into state for loss
                 loss_dt = self.loss_dt(p)
@@ -156,7 +157,7 @@ def __init__(self, p: Parameters):
 
         return None
 
-    def run_projection(self, p: Parameters, doubling_time: float) -> Tuple[pd.DataFrame, float]:
+    def run_projection_dt(self, p: Parameters, doubling_time: float) -> Tuple[pd.DataFrame, float]:
         intrinsic_growth_rate = self._intrinsic_growth_rate(doubling_time)
 
         recovery_days = p.recovery_days
@@ -207,8 +208,18 @@ def loss_dt(self, p: Parameters) -> float:
         return (p.current_hospitalized - predicted_current_hospitalized) ** 2
 
     def loss_ds(self, p: Parameters) -> float:
-        pass
-    
+        """loss with respect to days_since """
+
+        ## get the predicted number hospitalized today
+        pred_current_hospitalized = self.census_df['hospitalized'].loc[p.n_days_since_first_hospitalized]
+
+        ## compare against the actual (user inputed) number
+        ## squared difference is the loss to be optimized
+        return (p.current_hospitalized - pred_current_hospitalized)**2
+
+    def argmin_ds(self, p: Parameters) -> float:
+        losses_df = (self.census_df.hospitalized - p.current_hospitalized) ** 2
+        return losses_df.argmin()
 
     @staticmethod
     def _intrinsic_growth_rate(doubling_time: Optional[float]) -> float:
@@ -264,10 +275,11 @@ def sim_sir_df(
     s: float, i: float, r: float, beta: float, gamma: float, n_days: int
 ) -> pd.DataFrame:
     """Simulate the SIR model forward in time."""
-    return pd.DataFrame(
+    dat = pd.DataFrame(
         data=gen_sir(s, i, r, beta, gamma, n_days),
         columns=("day", "susceptible", "infected", "recovered"),
     )
+    return dat
 
 
 def get_dispositions(

src/penn_chime/presentation.py

@@ -497,6 +497,8 @@ def draw_census_table(
 def draw_raw_sir_simulation_table(st, model, parameters):
     as_date = parameters.as_date
     projection_area = model.raw_df
+    if parameters.n_days_since_first_hospitalized is not None:
+        projection_area.loc[:, "day"] = projection_area.day - parameters.n_days_since_first_hospitalized
     infect_table = (projection_area.iloc[::7, :]).apply(np.floor)
     infect_table.index = range(infect_table.shape[0])
     infect_table["day"] = infect_table.day.astype(int)

src/penn_chime/utils.py

@@ -53,11 +53,17 @@ def add_date_column(
     if date_first_hospitalized:
         delta = today - date_first_hospitalized
         start = today - delta
+        delta_days = delta.days
     else:
         start = today
-    end = start + timedelta(days=n_days + 1) # + delta # the +delta part is a hypothesis by phil
+        delta_days = 0
+        delta = timedelta(days=0)
+    end = start + timedelta(days=n_days + 1) + delta
+
     # And pick dates present in frame
-    dates = pd.date_range(start=start, end=end, freq="D")[df.day.tolist()]
+    dates = pd.date_range(
+        start=start, end=end, freq="D", closed="right"
+    )[[x + delta_days for x in df.day.tolist()]]
 
     if date_format is not None:
         dates = dates.strftime(date_format)
@@ -72,7 +78,6 @@ def add_date_column(
 
     # sort columns
     df = df[date_columns + non_date_columns]
-
     return df
 
 def dataframe_to_base64(df: pd.DataFrame) -> str: