Lowered knn default radius, removed interpolation from seasonal graphs.

main.py

@@ -2,6 +2,8 @@
 import math
 from typing import List, Tuple
 
+import plotly.graph_objects as go
+
 import pandas as pd
 
 from definitions import LATITUDES_BY_AIRPORT, LONGITUDES_BY_AIRPORT, POPULATION_DATA_DF, \
@@ -31,7 +33,7 @@ def get_nearest_city_from_airport(airport: str) -> Tuple[str, str]:
     return city, state
 
 
-def knn(city: str, state: str, k: int = 3, radius: int = 1) -> List[Tuple[str, str]]:
+def knn(city: str, state: str, k: int = 3, radius: int = .22) -> List[Tuple[str, str]]:
     """
     Gets the nearest k cities in POPULATION_DATA_DF from a city within the bounds of radius. Skips any cities in
     VISITED so that they are not double counted.
@@ -148,17 +150,19 @@ def main():
         'datetime64[ns, UTC]')
     population_weighted_min_temperatures_ts.index = population_weighted_min_temperatures_ts.index.astype(
         'datetime64[ns, UTC]')
-    p_21 = pd.concat([population_weighted_mean_temperatures_ts,
-                      population_weighted_mean_temperatures_ts.resample('3M').mean(),
-                      population_weighted_max_temperatures_ts.resample('3M').max(),
-                      population_weighted_min_temperatures_ts.resample('3M').min()], axis=1)
-
-    p_21.columns = ['Temperature', 'Seasonal Average', 'Seasonal Max', 'Seasonal Min']
-    p_21['Seasonal Average'].interpolate(method='time', inplace=True)
-    p_21['Seasonal Max'].interpolate(method='time', inplace=True)
-    p_21['Seasonal Min'].interpolate(method='time', inplace=True)
-
-    fig = p_21.plot(title="US Population Weighted Daily and Seasonal Temperature")
+    seasonal_mean = population_weighted_mean_temperatures_ts.resample('3M').mean()
+    seasonal_max = population_weighted_max_temperatures_ts.resample('3M').max()
+    seasonal_min = population_weighted_min_temperatures_ts.resample('3M').min()
+
+    data = [go.Scatter(x=population_weighted_mean_temperatures_ts.index,
+                       y=population_weighted_mean_temperatures_ts.values, name='Temperature'),
+            go.Scatter(x=seasonal_mean.index,
+                       y=seasonal_mean.values, name='Seasonal Average'),
+            go.Scatter(x=seasonal_max.index,
+                       y=seasonal_max.values, name='Seasonal Max'),
+            go.Scatter(x=seasonal_min.index,
+                       y=seasonal_min.values, name='Seasonal Min')]
+    fig = go.Figure(data=data)
     fig.show()
     fig.write_html(DAILY_SEASONAL_WEIGHTED_TEMPERATURE_GRAPH_PATH)
 

readme.txt

@@ -11,8 +11,11 @@ Washington and Wash DC/Dulles both refer to Washington DC.
 We are assuming the population of each city is a good weight to represent the average temperature in the US.
 For this exercise, I loop over each airport, interpolate missing data, find the closest city to the airport,
 take a simple coefficient of that city's population divided by the total population, and then do the same for k=3
-neighboring cities within approximately 69 miles. This is pretty arbitrary and only represents about 50% of the total
-US population. To account for more of the population, the temperature data would need to cover more land.
+neighboring cities within approximately 15 miles. This is pretty arbitrary and only represents about 40% of the total
+US population. To account for more of the population, the temperature data would need to cover more land. With the
+given dataset, there is a tradeoff between how much of the US population is represented, and accuracy of the representation.
+Weather is highly variable and can differ greatly over a small area, so the further away the city is from a station,
+the less accurate the representation will be.
 In reality, population weighted daily temperature timeseries is calculated in a more regional way
 (see https://www.eia.gov/outlooks/steo/special/pdf/2012_sp_04.pdf).