Setting frequency and debugging why nan values are being detected in train test split plus other model failures

[pranavlal]

Hi all,

I am trying to forecast stock prices. I have a pandas data frame containing ohlc data.

Several models however are detecting nan values in train test split and though the data is hourly, the autots library is unable to determine the frequency.

I am happy to specify the frequency but cannot determine how to do so. Yes, a datetime stamp is involved but what do I need to do? The data is hourly which is probably what I need to specify.

I am attaching a partial log of the run. My code is below.

[code]
import numpy as np
import pandas as pd
import sys
from datetime import datetime
from autots import AutoTS

path = r'/home/pranav/epat/stockbot/ohlc/hourly/SBIN.csv'

df=pd.read_csv(path,index_col="date",parse_dates=True)
print("frame description")
print(df.describe())
print("data frame info")
print(df.info())
model = AutoTS(forecast_length=6,frequency='infer',ensemble='all',max_generations=15,num_validations=4)
model=model.fit(df,)
prediction = model.predict()
print("model summary")
print(model)
model_results = model.results()
validation_results = model.results("validation")
print("validation results:")
print(validation_results)
[/code]

automatic_ts.txt

[winedarksea]

The way to set frequency is passing frequency="H" to the AutoTS() init where the string offset is one of the pandas recognized string offsets. You can also preset the frequency of the pandas index, I usually find .resample() works best for that.
It would be worth double-checking the model.df_wide_numeric of the AutoTS object after you .fit() to make sure it looks like you are expecting it to.