COVID19 California Data + Facebook Prophet¶
In [1]:
import pandas as pd
from fbprophet import Prophet
df = pd.read_csv('https://raw.githubusercontent.com/danielcaraway/data/master/ca_train.csv')
df.head()
Out[1]:
In [0]:
sm = df[['Date', 'ConfirmedCases']]
sm.columns = ['ds','y']
In [7]:
m = Prophet()
m.fit(sm)
Out[7]:
In [16]:
future = m.make_future_dataframe(periods=365)
future.tail()
Out[16]:
In [17]:
forecast = m.predict(future)
fig1 = m.plot(forecast)
* NOTE: Wow that is ridiculously unhelpful. Clearly 365 was too much. Turning it into a function for more rapid testing¶
OUR FUNCTION¶
In [0]:
def get_prof_preds_for(df, n):
m = Prophet()
m.fit(df)
future = m.make_future_dataframe(periods=n)
forecast = m.predict(future)
fig1 = m.plot(forecast)
In [19]:
get_prof_preds_for(sm, 60)
In [20]:
get_prof_preds_for(sm, 30)
In [21]:
get_prof_preds_for(sm, 10)
Trimming the training data¶
In [50]:
sm
Out[50]:
In [25]:
get_prof_preds_for(sm[20:],10)
In [0]:
def get_prof_preds_for(df, n):
m = Prophet(daily_seasonality=True)
m.fit(df)
future = m.make_future_dataframe(periods=n)
forecast = m.predict(future)
fig1 = m.plot(forecast)
In [30]:
get_prof_preds_for(sm[20:],10)
In [33]:
get_prof_preds_for(sm[40:],10)
OK looks like we are getting somewhere¶
In [35]:
get_prof_preds_for(sm[40:],60)
In [39]:
get_prof_preds_for(sm[47:],30]
In [44]:
get_prof_preds_for(sm[47:],20)
STEP 1B: Try with Fatalities¶
In [0]:
sm = df[['Date', 'Fatalities']]
sm.columns = ['ds','y']
In [46]:
get_prof_preds_for(sm, 20)
In [47]:
sm
Out[47]:
In [48]:
get_prof_preds_for(sm[47:], 20)
In [0]:
STEP 2: Actually make predictions¶
(using the mess we made above for fodder)
In [52]:
df.head()
Out[52]:
In [0]:
cc = df[['Date', 'ConfirmedCases']] # cc = Confirmed Cases
fc = df[['Date', 'Fatalities']] # fc = Fatal Cases
Make our function return something useful for a big df¶
In [0]:
def get_prof_preds_for(df, n):
df.columns = ['ds', 'y']
m = Prophet()
m.fit(df)
future = m.make_future_dataframe(periods=n)
forecast = m.predict(future)
fig1 = m.plot(forecast)
return forecast
In [63]:
pred_cc_1 = get_prof_preds_for(cc, 40)
pred_fc_1 = get_prof_preds_for(fc, 40)
In [66]:
pred_cc_1.tail()['yhat']
Out[66]:
In [68]:
pred_fc_1.tail()['trend']
Out[68]:
In [69]:
pred_fc_1.tail()
Out[69]:
In [71]:
pred_cc_2 = get_prof_preds_for(cc[40:], 40)
pred_fc_2 = get_prof_preds_for(fc[40:], 40)
In [72]:
pred_cc_2.tail()['yhat']
Out[72]:
In [73]:
pred_fc_2.tail()['yhat']
Out[73]:
In [75]:
pred_fc_2.tail()
Out[75]:
In [76]:
pred_cc_3 = get_prof_preds_for(cc[47:], 40)
pred_fc_3 = get_prof_preds_for(fc[47:], 40)
In [77]:
pred_cc_3.tail()
Out[77]:
In [78]:
pred_fc_3.tail()
Out[78]:
In [0]:
pred_cc_1_sm = pred_cc_1[['ds','yhat']]
pred_cc_2_sm = pred_cc_2[['ds','yhat']]
pred_cc_3_sm = pred_cc_3[['ds','yhat']]
In [0]:
# lol pandas says this way
# pred_cc_1_sm.merge(pred_cc_2_sm, left_on='ds')
# stackoverflow says this way (better)
cc_df = pd.merge(pred_cc_1_sm, pred_cc_2_sm, how='left', on='ds', suffixes=('_v1', '_v2'))
cc_df = pd.merge(cc_df, pred_cc_3_sm, how='left', on='ds')
In [139]:
cc_df.columns = ['ds','yhat_v1', 'yhat_v2', 'yhat_v3']
cc_df
Out[139]:
In [0]:
pred_fc_1_sm = pred_fc_1[['ds','yhat']]
pred_fc_2_sm = pred_fc_2[['ds','yhat']]
pred_fc_3_sm = pred_fc_3[['ds','yhat']]
fc_df = pd.merge(pred_fc_1_sm, pred_fc_2_sm, how='left', on='ds', suffixes=('_v1', '_v2'))
fc_df = pd.merge(fc_df, pred_fc_3_sm, how='left', on='ds')
In [104]:
fc_df.columns = ['ds','yhat_v1', 'yhat_v2', 'yhat_v3']
fc_df
Out[104]:
In [0]:
import matplotlib.pyplot as plt
In [118]:
cc_df['datetime'] = pd.to_numeric(cc_df['ds'])
# cc_df = cc_df.set_index('datetime')
cc_df.plot(x="datetime", y=["yhat_v1", "yhat_v2", "yhat_v3"])
plt.show()
In [124]:
cc_df.dtypes
Out[124]:
In [133]:
import seaborn as sns
# sns.lineplot()
# ax = sns.lineplot(x="ds", y="yhat_v1", data=cc_df);
ax = sns.lineplot(x="ds", y="yhat_v1", data=cc_df)
In [129]:
ax = sns.lineplot(x="ds", y="yhat_v1", data=cc_df)
In [0]:
fmri = sns.load_dataset("fmri")
In [135]:
fmri.head()
Out[135]:
In [140]:
cc_df
Out[140]:
In [0]:
cc_df_melt = pd.melt(cc_df, id_vars=['ds'], value_vars=['yhat_v1', 'yhat_v2', 'yhat_v3'])
In [150]:
cc_df_melt
Out[150]:
In [151]:
sns.relplot(x="ds", y="value", hue="variable", style="variable",
kind="line", data=cc_df_melt);
In [0]: