Apply ta_py function to Cudf dataframe - RAPIDS

Question

trying to create a new column on a cudf dataframe based on VWMA from ta_py :

#creating df
CJ_m30 = cudf.read_csv("/media/f333a/Data/CJ_m30.csv", 
                         names = ["DateTime","Bid","Ask","Open", "High", "Low", "Close"])

#trying to create new column based on func

import ta_py as ta

length = 40

def process_vwma(data):
    VWMA = ta.vwma(data,length)
    return  VWMA
    

CJ_m30['VWMA'] = CJ_m30['Close'].apply(process_vwma, axis = 0)

returns error :

ValueError: UDFs using *args or **kwargs are not yet supported.

updated: now error is :

TypingError: Failed in cuda mode pipeline (step: nopython frontend) Failed in cuda mode pipeline (step: nopython frontend) Unknown attribute 'vwma' of type Module(<module 'ta_py' from '/home/f320x/anaconda3/envs/rapids-22.02/lib/python3.9/site-packages/ta_py/init.py'>)

File "../../../../../tmp/ipykernel_3478/1395824149.py", line 6:

During: typing of get attribute at /tmp/ipykernel_3478/1395824149.py (6)

File "../../../../../tmp/ipykernel_3478/1395824149.py", line 6:

During: resolving callee type: type(<numba.cuda.compiler.Dispatcher object at 0x7f05b67a47c0>) During: typing of call at /home/f320x/anaconda3/envs/rapids-22.02/lib/python3.9/site-packages/cudf/core/series.py (2495)

File "../../../anaconda3/envs/rapids-22.02/lib/python3.9/site-packages/cudf/core/series.py", line 2495: >>> def f(x):
    return df.apply(lambda row: f_(row[name]))
    ^

Can someone give an explanation ? Thank you

You can't run arbitrary functions with cuDF's `apply`. More importantly, if your function expects a pandas DataFrame you may not be able to pass a cuDF DataFrame to it. To run such a function, you may either need to convert to a pandas DataFrame or rewrite your function to accept cuDF inputs. — Nick Becker, Apr 16 '22 at 16:38
I see you've had a few questions around this. You should join the NVIDIA Developer Forums (https://forums.developer.nvidia.com/) or on Slack at the RAPIDS GoAI Channel (https://join.slack.com/t/rapids-goai/shared_invite/zt-trnsul8g-Sblci8dk6dIoEeGpoFcFOQ). There is an FSI team we can introduce you to. who may be able to help you with these industry use cases — TaureanDyerNV, May 02 '22 at 19:49

score 0 · Answer 1 · answered May 02 '22 at 19:32

Building on Nick's comment, the data used in VWMA in ta_py, https://github.com/Bitvested/ta.py#vwma, is structured very differently than you have it as or are passing into it. It's an array with price (close), but also volume, which you don't have. The data you're giving it is for is SMA - but not in the way the functions should be used, as Nick said.

If you want to use Close for SMA, borrowing from https://www.datacamp.com/community/tutorials/moving-averages-in-pandas

import cudf
product = {'day' : [1,2,3,4,5,6,7,8,9,10,11,12],'close':[290,260,288,300,310,303,329,340,316,330,308,310]}
df = cudf.DataFrame(product)
print(df.head())
df['SMA'] = df.iloc[:,1].rolling(window=3).mean() # you change your window size , eg from "3" in this example to "40"
print(df.head())

which outputs this:

    day close   SMA
0   1   290 <NA>
1   2   260 <NA>
2   3   288 279.3333333
3   4   300 282.6666667
4   5   310 299.3333333

If you want VWMA, get a dataset that has volume data as well. This keeps everything on GPU, instead of mixing CPU and GPU libraries. Also look into cuSignal for signal analysis.

Apply ta_py function to Cudf dataframe - RAPIDS

1 Answers1