Build an algorithm

Basically I’m looking to;
• upload BTC timeseries data (.csv) – this can be by the minute or it can be daily or indeed weekly
• perform some analysis
• date time feature – possibly convert into ‘year’ ‘month’ ‘day’
• transform data to eliminate trends (if they exist)
• EDA
• create/use base model & determine RMSE
• create a lag feature
• visualise the data
o lag scatter plots
o autocorrelation plots etc
• moving average (& as prediction or for comparison)
• check for stationarity
• evaluate models (backtesting)
• reframe timeseries (create lagged dataset & classification function to predict output) ie;
# Create lagged dataset
values = pd.DataFrame(s.values)
df = pd.concat([values.shift(1), values], axis=1)
df.columns = ['t', 't+1']

def make_discrete(row):
if row['t+1'] 3:
return 'high'
else:
return 'medium'

# apply the above function to reassign t+1 values
df['t+1'] = df.apply(lambda row: make_discrete(row), axis=1)

• improve RMSE using new fine tuned models
• compare autoregression/Arima/etc models
• grid search model hyperparameters
• evaluate forecasts
• compare and display models