COVID-19 analysis code 05 - plot predictions

covid19_analysis_05_plot_predictions.py

# Imports for performing ML analysis
from sklearn.linear_model import LinearRegression

from scipy.optimize import curve_fit

from datetime import timedelta

# Set range of data to build model
# It might make sense to skip part of the initial points, when the exponential trend was still not evident
START_DATE = datetime(2020, 2, 23).date()

# Prepare value vectors

# The acual values
y = country_df[country_df['Date'] >= START_DATE]['Confirmed']

# The log values
y_log = np.log(y)

# Independent variable
x = np.arange(len(y))

# The two versions, weighted and unweighted
reg_unweighted = LinearRegression()
reg_unweighted.fit(x[:,np.newaxis], y_log)

reg_weighted = LinearRegression()
reg_weighted.fit(x[:,np.newaxis], y_log, sample_weight=y)

# Create a dataframe with predicted values
PREDICT_UNTIL = (datetime.today() + timedelta(days=1)).date().strftime("%m/%d/%Y")

# Prepare range of dates
estimate_dates = pd.date_range(start=START_DATE.strftime("%m/%d/%Y"), end=PREDICT_UNTIL)

# Make predictions
estimate_cases_ols_unweighted = np.exp(reg_unweighted.predict(np.arange(len(estimate_dates))[:, np.newaxis]))
estimate_cases_ols_weighted = np.exp(reg_weighted.predict(np.arange(len(estimate_dates))[:, np.newaxis]))

# df_estimates = pd.DataFrame({"Date": estimate_dates, "Predictions": estimate_cases})
df_estimates = pd.DataFrame({
  "Date": estimate_dates, 
  "Predictions (unweighted)": estimate_cases_ols_unweighted, 
  "Predictions (weighted)": estimate_cases_ols_weighted})

# Plot values
ax = plt.gca()

country_df.plot(
    x='Date', 
    y=["Confirmed"], 
    figsize=(20,10), ax=ax, marker='o')

# Uncomment "Predictions (unweighted)" line to show also the plot relative to 
# (since its values are much higher it woudld compress the other curves)
df_estimates.plot(
    x='Date', 
    y=[
#         "Predictions (unweighted)",
        "Predictions (weighted)"
    ], 
    figsize=(20,10), ax=ax, marker='o', alpha=0.4, color=['green', 'orange'])

ax.xaxis.set_major_locator(ticker.MultipleLocator(3))

ax.set_ylabel("# of confirmed cases");

# # Zoom in
# ax.set_xlim(["2020-02-20", "2020-03-12"]);