|
|
from darts import TimeSeries |
|
|
from darts.datasets import ILINetDataset |
|
|
from darts.metrics import mape |
|
|
from darts.models import ExponentialSmoothing |
|
|
from darts.utils.missing_values import fill_missing_values |
|
|
from darts.dataprocessing.transformers import Scaler |
|
|
import matplotlib.pyplot as plt |
|
|
import numpy as np |
|
|
import pandas as pd |
|
|
import os |
|
|
|
|
|
def load_ILINetDataset(): |
|
|
""" |
|
|
Dataset's Components Descriptions: |
|
|
|
|
|
* % WEIGHTED ILI: Combined state-specific data of patients visit to healthcare providers for ILI reported each week weighted by state population |
|
|
|
|
|
* % UNWEIGHTED ILI: Combined state-specific data of patients visit to healthcare providers for ILI reported each week unweighted by state population |
|
|
|
|
|
* AGE 0-4: Number of patients between 0 and 4 years of age |
|
|
|
|
|
* AGE 25-49: Number of patients between 25 and 49 years of age |
|
|
|
|
|
* AGE 25-64: Number of patients between 25 and 64 years of age |
|
|
|
|
|
* AGE 5-24: Number of patients between 5 and 24 years of age |
|
|
|
|
|
* AGE 50-64: Number of patients between 50 and 64 years of age |
|
|
|
|
|
* AGE 65: Number of patients above (>=65) 65 years of age |
|
|
|
|
|
* ILITOTAL: Total number of ILI patients. For this system, ILI is defined as fever (temperature of 100°F [37.8°C] or greater) and a cough and/or a sore throat |
|
|
|
|
|
* NUM. OF PROVIDERS: Number of outpatient healthcare providers |
|
|
|
|
|
* TOTAL PATIENTS: Total number of patients |
|
|
""" |
|
|
ilidata = ILINetDataset().load().astype(np.float32) |
|
|
|
|
|
return ilidata |
|
|
|
|
|
|
|
|
def preprocess_data(ilidata): |
|
|
ilitotal = ilidata['ILITOTAL'] |
|
|
covariates = ilidata.drop_columns(col_names='ILITOTAL') |
|
|
|
|
|
|
|
|
log_ili = ilitotal.map(np.log) |
|
|
|
|
|
|
|
|
pd_log_ili = log_ili.pd_dataframe() |
|
|
pd_log_ili['ILITOTAL'].replace(to_replace=-np.inf, value=np.nan, inplace=True) |
|
|
log_ili = TimeSeries.from_dataframe(pd_log_ili) |
|
|
|
|
|
log_ili = fill_missing_values(log_ili) |
|
|
|
|
|
return log_ili |
|
|
|
|
|
|
|
|
def train_val_split(ilidata): |
|
|
train_ili, val_ili = ilidata.split_before(0.75) |
|
|
|
|
|
return train_ili, val_ili |
|
|
|
|
|
|
|
|
def scale_train(train_ili): |
|
|
scaler = Scaler() |
|
|
train_ili_scaled = scaler.fit_transform(train_ili) |
|
|
|
|
|
return train_ili_scaled, scaler |
|
|
|
|
|
|
|
|
def train(train_ili_scaled): |
|
|
model = ExponentialSmoothing() |
|
|
model.fit(train_ili_scaled) |
|
|
|
|
|
return model |
|
|
|
|
|
|
|
|
def save_model(model, path): |
|
|
model_name = str(model).split('(')[0] |
|
|
if not os.path.exists(path=path): |
|
|
os.makedirs(path) |
|
|
model.save(os.path.join(path, model_name)) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def load_model(path): |
|
|
model = ExponentialSmoothing.load(path) |
|
|
|
|
|
return model |
|
|
|
|
|
|
|
|
def predict(model, val_ili): |
|
|
preds = model.predict(len(val_ili)) |
|
|
|
|
|
return preds |
|
|
|
|
|
|
|
|
def inverse_scale_predictions(scaled_preds, scaler): |
|
|
preds = scaler.inverse_transform(scaled_preds) |
|
|
|
|
|
return preds |
|
|
|
|
|
|
|
|
def plot_results(train_ili, val_ili, preds): |
|
|
fig, _ = plt.subplots() |
|
|
train_ili.plot(label='train') |
|
|
val_ili.plot(label='val') |
|
|
preds.plot(label='preds') |
|
|
|
|
|
return fig |
|
|
|
|
|
|
|
|
def compute_mape(preds, val): |
|
|
metric = mape(val, preds) |
|
|
|
|
|
return metric |
|
|
|
|
|
|
