# Sub-daily data¶

NeuralProphet can make forecasts for time series with sub-daily observations by passing in a dataframe with timestamps in the ds column. The format of the timestamps should be YYYY-MM-DD HH:MM:SS - see the example csv here. When sub-daily data are used, daily seasonality will automatically be fit.

Here we fit NeuralProphet to data with 5-minute resolution (daily temperatures at Yosemite).

[1]:

if 'google.colab' in str(get_ipython()):
!pip install git+https://github.com/ourownstory/neural_prophet.git # may take a while
#!pip install neuralprophet # much faster, but may not have the latest upgrades/bugfixes

import pandas as pd
from neuralprophet import NeuralProphet, set_log_level
set_log_level("ERROR")

[2]:

data_location = "https://raw.githubusercontent.com/ourownstory/neuralprophet-data/main/datasets/"



Now we will attempt to forecast the next 7 days. The 5min data resulution means that we have 60/5*24=288 daily values. Thus, we want to forecast 7*288 periods ahead.

Using some common sense, we set: * First, we disable weekly seasonality, as nature does not follow the human week’s calendar. * Second, we disable changepoints, as the dataset only contains two months of data

[3]:

m = NeuralProphet(
n_changepoints=0,
weekly_seasonality=False,
)
metrics = m.fit(df, freq='5min')
forecast = m.predict(df)
fig = m.plot(forecast)
# fig_comp = m.plot_components(forecast)
fig_param = m.plot_parameters()




The daily seasonality seems to make sense, when we account for the time being recorded in GMT, while Yosemite local time is GMT-8.

## Improving trend and seasonality¶

As we have 288 daily values recorded, we can increase the flexibility of daily_seasonality, without danger of overfitting.

Further, we may want to re-visit our decision to disable changepoints, as the data clearly shows changes in trend, as is typical with the weather. We make the following changes: * increase the changepoints_range, as the we are doing a short-term prediction * inrease the n_changepoints to allow to fit to the sudden changes in trend * carefully regularize the trend changepoints by setting trend_reg in order to avoid overfitting

[6]:

m = NeuralProphet(
changepoints_range=0.95,
n_changepoints=50,
trend_reg=1,
weekly_seasonality=False,
daily_seasonality=10,
)
metrics = m.fit(df, freq='5min')
forecast = m.predict(df)
fig = m.plot(forecast)
# fig_comp = m.plot_components(forecast)
fig_param = m.plot_parameters()



[ ]: