Getting more values than specified in predictions

[pranavlal]

Hi all,

I have specified a prediction length of 22 for my time series but I am getting over a thousand values. I suspect I am using the model incorrectly. How do I correct my mistake? Here is the relevant portion of my code.

I am following the tutorial at the below link.

def load_model():
	pipeline = ChronosPipeline.from_pretrained("amazon/chronos-t5-base",device_map="cuda",torch_dtype=torch.bfloat16,)
	return pipeline

def predict(pipeline, timeseries, prediction_length=12):
	"""given a timeseries, predict with chronos model"""
	forecast = pipeline.predict(timeseries, prediction_length, num_samples=1)[0][0]  # shape [num_series, num_samples, prediction_length]
	return forecast

ts = torch.tensor(df["returns"].to_numpy(), dtype=torch.float32)
	pipeline = load_model()
	forecast = predict(pipeline, ts,64)
	npa=ts.detach().tolist()
	print(len(npa))
	

[lostella]

Hi @pranavlal, it looks like here you're printing the size of the input series, not of the forecast

npa=ts.detach().tolist()
print(len(npa))

[pranavlal]

Hi @ Lorenzo, Thanks, I wanted to count the number of entries in the forecast to determine if I was getting the expected number of entries. From: Lorenzo Stella ***@***.***> Sent: Monday, April 29, 2024 12:49 PM To: amazon-science/chronos-forecasting ***@***.***> Cc: Pranav Lal ***@***.***>; Mention ***@***.***> Subject: Re: [amazon-science/chronos-forecasting] Getting more values than specified in predictions (Discussion #59) Hi @pranavlal <https://github.com/pranavlal> , it looks like here you're printing the size of the input series, not of the forecast npa=ts.detach().tolist() print(len(npa)) — Reply to this email directly, view it on GitHub <#59 (comment)> , or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACCF2A4COWAIX4LCNWTPVJLY7XX7LAVCNFSM6AAAAABG5MGSHWVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TENJYGU4TK> . You are receiving this because you were mentioned. <https://github.com/notifications/beacon/ACCF2A3BKX4WC5YJ5JTNBNTY7XX7LA5CNFSM6AAAAABG5MGSHWWGG33NNVSW45C7OR4XAZNRIRUXGY3VONZWS33OINXW23LFNZ2KUY3PNVWWK3TUL5UWJTQARVDGG.gif> Message ID: ***@***.*** ***@***.***> >

[lostella]

For that you should do

print(forecast.shape)

in your code above