ADX raises TypeError

Hey,

I was trying to use the ADX (Average Directional Index) Momentum Indicator.
For me this raises a TypeError:

import talib
from talib import abstract

adx = abstract.ADX
result = adx(dataset, time_period=14) # <-- TypeError here

The full stack trace:

TypeError                                 Traceback (most recent call last)
<ipython-input-90-36d2703ed47a> in <module>()
      2 
      3 adx = tabst.ADX
----> 4 result = adx(dataset, time_period=14)

talib/_abstract.pxi in talib._ta_lib.Function.__call__()

talib/_abstract.pxi in talib._ta_lib.Function.__call_function()

~/.local/share/virtualenvs/project/lib/python3.5/site-packages/talib/__init__.py in wrapper(*args, **kwargs)
     30                       for k, v in kwargs.items()}
     31 
---> 32             result = func(*args, **kwargs)
     33 
     34             # Series was passed in, Series gets out; re-apply index

talib/_func.pxi in talib._ta_lib.ADX()

TypeError: only size-1 arrays can be converted to Python scalars

I have tested with both pandas.DataFrame and a dictionary of numpy.ndarrays. Result is the same.

The other functions work fine though, this one seems to be the only one causing this behaviour for me.

I built TA Lib from source using the instructions provided. Running on Ubuntu 16.04 LTS.

Regards.

What is dataset?

It works for me:

>>> import talib
>>> import talib.abstract
>>> import numpy

>>> high = numpy.random.randn(50)
>>> low = numpy.random.randn(50)
>>> close = numpy.random.randn(50)

>>> talib.ADX(high, low, close, timeperiod=14)

>>> dataset = {'high': high, 'low': low, 'close': close}
>>> talib.abstract.ADX(dataset, timeperiod=14)

I tried your code and it works.. I did some further investigation and even though this does not produce the exact same error it will cause a similar one:

import numpy as np
import pandas as pd
import talib
from talib import abstract as tabst

d = { i: np.random.randn(50) for i in [ 'high', 'low', 'close' ] }
adx = tabst.ADX
try:
    result = adx(d['high'], d['low'], d['close'], timeperiod=14)
except TypeError:
    pass

result = adx(d) # <-- TypeError here

The stack trace of the error:

---------------------------------------------------------------------------
StopIteration                             Traceback (most recent call last)
~/.local/share/virtualenvs/project/lib/python3.5/site-packages/talib/__init__.py in wrapper(*args, **kwargs)
     19                 index = next(arg.index
---> 20                              for arg in chain(args, kwargs.values())
     21                              if isinstance(arg, _pd_Series))

StopIteration: 

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
<ipython-input-3-3aa40928cfa5> in <module>()
----> 1 result = adx(d)

talib/_abstract.pxi in talib._ta_lib.Function.__call__()

talib/_abstract.pxi in talib._ta_lib.Function.__call_function()

~/.local/share/virtualenvs/project/lib/python3.5/site-packages/talib/__init__.py in wrapper(*args, **kwargs)
     22             except StopIteration:
     23                 # No pandas.Series passed in; short-circuit
---> 24                 return func(*args, **kwargs)
     25 
     26             # Use Series' float64 values if pandas, else use values as passed

talib/_func.pxi in talib._ta_lib.ADX()

TypeError: only size-1 arrays can be converted to Python scalars

Apparently it was caused by me experimenting with the api and calling the function with wrong arguments (I was using a jupyter notebook)

Hmm, so I can replicate your problem and will look into it.

But, adding timeperiod as a keyword argument "fixes" it?

adx(d, timeperiod=14)

It seems to be caused by calling a function with the wrong parameters followed by a call which uses the default parameters.
Adding all required parameters explicitly seems to "fix" it.

@briancappello

@briancappello any interest in looking into this bug with abstract wrapper?

I guess I'm not clear on what the error is? I'm able to reproduce the exception by calling abstract.ADX(d['high'], d['low'], d['close']), but if I remember correctly, I think all the abstract functions expect the first argument to be a dictionary/dataframe of prices (as opposed to passing the price arrays as individual arguments).

import numpy as np
import talib
from talib import abstract

d = {i: np.random.randn(50) for i in ['open', 'high', 'low', 'close']}

func_result = talib.ADX(d['high'], d['low'], d['close'])
abstract_result = abstract.ADX(d)
assert func_result.all() == abstract_result.all()

func_result = talib.ADX(d['high'], d['low'], d['close'], timeperiod=20)
abstract_result = abstract.ADX(d, timeperiod=20)
assert func_result.all() == abstract_result.all()

func_result = talib.ADX(d['high'], d['low'], d['open'])
abstract_result = abstract.ADX(d, prices=['high', 'low', 'open'])
assert func_result.all() == abstract_result.all()

Seems like errors cause future calls to not work. I modified your code to have a bad initial call, that catches and ignores the exception, then the later calls don't work.

import numpy as np
import pandas as pd
import talib
from talib import abstract

d = {i: np.random.randn(50) for i in ['open', 'high', 'low', 'close']}

try:
    func_result = abstract.ADX(d['high'], d['low'], d['close'])
except:
    pass

func_result = talib.ADX(d['high'], d['low'], d['close'])
abstract_result = abstract.ADX(d)
assert func_result.all() == abstract_result.all()

func_result = talib.ADX(d['high'], d['low'], d['close'], timeperiod=20)
abstract_result = abstract.ADX(d, timeperiod=20)
assert func_result.all() == abstract_result.all()

func_result = talib.ADX(d['high'], d['low'], d['open'])
abstract_result = abstract.ADX(d, prices=['high', 'low', 'open'])
assert func_result.all() == abstract_result.all()

Gotchya. Skipped right over this:

It seems to be caused by calling a function with the wrong parameters followed by a call which uses the default parameters.

Certainly I agree it's not optimal. Looking through the code in Function.set_function_args it looks like it's setting the function parameters positionally using the arrays (in this case, I think it's setting timeperiod to be d['low']). (in other words, abstract.ADX(d, 20) is the same as abstract.ADX(d, timeperiod=20). I suppose i could put a type check in there that if the value for the function parameters isn't a "simple type" (str/bool/int/float) don't set it? In hind sight it would have been better to make those function parameters kwargs-only.

Maybe the problem is saving keyword arguments?

import numpy as np
import talib
from talib import abstract

d = {i: np.random.randn(50) for i in ['open', 'high', 'low', 'close']}

result1 = abstract.ADX(d, timeperiod=5)

result2 = abstract.ADX(d)

assert result1.all() == result2.all()

Seems weird to me that the second call uses timeperiod=5 instead of the timeperiod=DEFAULT?

I mean, yea. That was a poor design decision on my part. And I'd be OK with changing it, but technically I think it would break BC. Which might not that big of a deal in this case - what do you think?

I'm not actually sure that would work. It could break a fair number of things, eg if you needed to determine how much data you needed based on the lookback:

adx20 = abstract.ADX
adx20.set_parameters({'timeperiod': 20})
d = {i: np.random.randn(50 + adx20.lookback) for i in ['high', 'low', 'close']}
adx20(d)

The only thing I can think of at the moment that wouldn't break BC and would fix this is the hacky type check. Second best might be to only accept prices as a positional argument to __call__, but doing that could break code in the wild. Ultimately, and I admit the public interface/documentation perhaps isn't the greatest, but be that as it may, it seems like maybe this is really closer to new-user-error than a bug per se?

Hmm, that's an excellent point.

I'm not sure I'd say it's a design flaw. Seems like caching the parameters works nicely with lookback. Although we could have another version where lookback takes the parameters as function arguments, so no caching.

I'm not sure which way to proceed. It's hard to know if breaking backwards compatibility would impact anyone, or if we could change it in such a way where we preserved it if set_parameters called, but did not save the arguments if the __call__ was made. Separating intention from accident. Dunno.

Or we could just document it, and add a test case showing the exception clearing up next time you call it "correctly":

import numpy as np
import pandas as pd
import talib
from talib import abstract

d = {i: np.random.randn(50) for i in ['open', 'high', 'low', 'close']}

try:
    result1 = abstract.ADX(d, timeperiod="asdf")
except:
    pass # assertRaises

try:
    result2 = abstract.ADX(d)
except:
    pass # assertRaises

result3 = abstract.ADX(d, timeperiod=14)

result4 = abstract.ADX(d)

assert result3.all() == result4.all()

The original reports were fixed by #195. One path is still open: a call that fails part-way keeps the parameters it was given, so the next default call fails too (SMA(d, timeperiod=0) then SMA(d)). #779 restores them on every path.

Navigation Menu

Repository navigation

ADX raises TypeError #194

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

ADX raises TypeError #194

Description

Activity

mrjbq7 commented on Apr 5, 2018

liona24 commented on Apr 6, 2018

mrjbq7 commented on Apr 6, 2018

liona24 commented on Apr 6, 2018

mrjbq7 commented on Apr 6, 2018

briancappello commented on Apr 6, 2018

mrjbq7 commented on Apr 6, 2018

briancappello commented on Apr 6, 2018

mrjbq7 commented on Apr 6, 2018

briancappello commented on Apr 6, 2018

briancappello commented on Apr 6, 2018

mrjbq7 commented on Apr 7, 2018

mrjbq7 commented on Apr 7, 2018

mario4tier commented on Oct 23, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions