numexpr
numexpr copied to clipboard
pydata
Fast numerical array expression evaluator for Python, NumPy, Pandas, PyTables and more
Hey! it looks like the reduce opcodes `('sum', 0, 'f')` and `('max', 0, 'f')` aren't present in `ne3.ne3compiler.OPTABLE` So the expression `max(x)` where x is of shape `(1024,)` only is...
Python will be able to handle nanoseconds when [this](https://bugs.python.org/issue15443) is solved. When it is, we should see nanoseconds specified [here](https://docs.python.org/3/library/datetime.html). Pandas can already handle [nanoseconds](https://pandas.pydata.org/pandas-docs/stable/generated/pandas.Series.dt.nanosecond.html). It also looks like numpy...
Hello, is there any support (or any plans for that matter) to not only have scalar outputs but for example vectors (either by specifying multiple expressions for each element or...
Trying to sum using boolean elements, result in an error ``` import numpy as np import numexpr as ne a = np.arange(10) ne.evaluate('sum(a>5)') ``` ``` KeyError: ('sum(a>5)', (('optimization', 'aggressive'), ('truediv',...
Yeppp! is a SIMD-based mathematics library from Georgia Tech that achieves similar performance to Intel’s VML, but Yeppp! comes under the permissive BSD license, the same as NE3. TO DO:...
In NE2 loops were unrolled, which resulted in a strong preference for a BLOCK_SIZE in terms of elements rather than bytes. This meant that the BLOCK_SIZE was generally optimized relative...
NE3 relies on the GNU-standard library for most mathematical functions. These functions are inlined, but in general there are mathematical libraries available that are a solid 2-4x faster. One such...
Consider the following simple example: ``` def calc_squares_sum(ch: bytearray, size: int): a = numpy.frombuffer(ch, dtype='uint8', count=size) a = a.astype(numpy.int64, copy=False) # Apparently it does copy anyway, when the type is...
Hello, I'm calculating the distance between all `rows` of `matrix m` and some `vector` `v`. m is a large matrix, about 500,000 rows and 2048 column. v is the size...
Numexpr would benefit from a flake8 plugin that checks code in `ne.evaluate` calls, and silences warnings about unused variables that are in fact used in numexpr code.
