search

sryza / spark-timeseries

A library for time series analysis on Apache Spark
Apache License 2.0
1.19k stars 424 forks source link

TypeError: 'JavaPackage' object is not callable #136

Open bertocast opened 8 years ago

bertocast commented 8 years ago

I'm getting this error when I try to create a new ARIMAModel.

from sparkts.models.ARIMA import autofit
from pyspark.context import SparkContext

a = [1, 2, 3 ,4 ,5]
model = autofit(a, sc=SparkContext('local', 'blabla'))

It's SparkContext fault, I think. Any thoughts?

sryza commented 8 years ago

@pegli any idea why this wouldn't be working?

pegli commented 8 years ago

My guess is that it is happening because the current version of ARIMA.autofit() is looking for a pyspark.mllib.linalg.Vector as the first parameter, not an array. The changes I made in PR #137 should fix this issue.

I tried to run the code above on my laptop (which has the code from the PR) and the call to autofit() does not result in a JavaError. However, it does appear to hang when provided with [1,2,3,4,5] as source data. When I increased the number of data points in the array, autofit() returned almost immediately. Here's my test code:

import numpy as np
from sparkts.models.ARIMA import autofit
from pyspark.context import SparkContext

a = np.linspace(1, 5, 10) # array starting at 1, ending at 5, and containing 10 elements
model = autofit(a, sc=SparkContext('local', 'blabla'))
print model.coefficients