can we reduce the prediction accuracy about statsforecast

[jmoralez]

Can you set the env variable before importing statsforecast?

[jmoralez]

You can set the NUMBA_DEBUG_CACHE env variable to see where the cache is being saved. Here's a minimal example:

import os

from numba import njit

os.environ['NUMBA_DEBUG_CACHE'] = '1'

@njit(cache=True)
def f(x):
    return x + 1

print(f(1))

running this script I get:

[cache] index saved to 'C:\\Users\\MyUser\\__pycache__\\test.f-7.py310.nbi'
[cache] data saved to 'C:\\Users\\MyUser\\__pycache__\\test.f-7.py310.1.nbc'
2

[jmoralez]

Hey @neonarc4, thanks for using statsforecast. What do you mean by "reduce the prediction accuracy"? If you want to reduce the compile time you can try caching and if that's the case please update the issue title.

[neonarc4]

os.environ["NIXTLA_NUMBA_CACHE"] = 1
os.environ["NUMBA_CACHE_DIR"] = 'numba_cache'
from statsforecast import StatsForecast
from statsforecast.models import AutoARIMA

sf = StatsForecast(
    models=[AutoARIMA(season_length = 12)],
    freq='M',
)

sf.fit(df)

is this correct approach ? on window 11

[neonarc4]

the folder remain blank for cache

[jmoralez]

Which version of statsforecast are you using?

[neonarc4]

alright now something broken is ur library doesnt support amd ? dml

[jmoralez]

statsforecast doesn't support GPU acceleration

[neonarc4]

ah lol then how can i fix this caching issue?

[jmoralez]

Which version of statsforecast are you using? The caching variable was introduced in 1.7.0.

[neonarc4]

@jmoralez 1.7.0

from statsforecast import StatsForecast
from statsforecast.models import AutoARIMA
import pandas as pd
import os

os.environ["NIXTLA_NUMBA_CACHE"] = "1"
os.environ["NUMBA_CACHE_DIR"] = "X:/sad/practice/test/aii/data"
df = pd.read_csv("https://datasets-nixtla.s3.amazonaws.com/air-passengers.csv", parse_dates=["ds"])

df.head()

sf = StatsForecast(
    models=[AutoARIMA(season_length=12)],
    freq="M",
)

sf.fit(df)
forecast_df = sf.predict(h=12, level=[90])
forecast_df.tail()
fig = sf.plot(df, forecast_df, level=[90])
fig.savefig("myfigure.png")

[neonarc4]

[jmoralez]

os.environ["NUMBA_CACHE_DIR"] = "X:/sad/practice/test/aii/data"

I don't think that's a valid path in Windows, it should probably be something like: "X:\sad\practice\test\aii\data"`. You can also not set this variable and just use the default directory (pycache)

[neonarc4]

no luck plz guide me bro

[neonarc4]

sorry fell sleep it doesnt work with statsforcecast

[neonarc4]

from statsforecast import StatsForecast
from statsforecast.models import AutoARIMA
import pandas as pd
import os
from tempfile import TemporaryDirectory
import time

# with TemporaryDirectory() as temp_dir:
#     os.environ["NUMBA_CACHE_DIR"] = temp_dir
os.environ["NIXTLA_NUMBA_CACHE"] = "1"
os.environ["NUMBA_DEBUG_CACHE"] = "1"
df = pd.read_csv("https://datasets-nixtla.s3.amazonaws.com/air-passengers.csv", parse_dates=["ds"])

df.head()

sf = StatsForecast(
    models=[AutoARIMA(season_length=12)],
    freq="M",
)

sf.fit(df)
forecast_df = sf.predict(h=12, level=[90])
forecast_df.tail()

fig = sf.plot(df, forecast_df, level=[90])
fig.savefig("myfigure.png")

ur code work but when i put it on statsforccast it doesnt

[neonarc4]

i love u bro amazing >.</ u fixed thnx alot may god bless u

[neonarc4]

between how it so fast without gpu need ? is my rx 7900 xtx is no needed ? or do i need to run linux machine for acceration on numba or ur library doesnt need gpu ? i am really curious how u able to achieve such insane speed without gpu kinda