The "validation" problems have invalid imports.

Specifically, line 11 of smoother_friedman82.py should be:

from .. import smoother

This is consisstent with both standards for Python 2.7 and Python 3.5

This is a problem with the README.rst documentation.

import matplotlib.pyplot as plt
%matplotlib inline
from ace import ace
ace.plot_transforms(myace, fname = 'mytransforms.pdf')

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-5-3ea400bca898> in <module>()
      2 get_ipython().magic(u'matplotlib inline')
      3 from ace import ace
----> 4 ace.plot_transforms(myace, fname = 'mytransforms.pdf')

/Users/paulperry/anaconda/lib/python2.7/site-packages/ace/ace.pyc in plot_transforms(ace_model, fname)
    241     plt.rcParams.update({'font.size': 8})
    242     plt.figure()
--> 243     numCols = len(ace_model.x) / 2 + 1
    244     for i in range(len(ace_model.x)):
    245         plt.subplot(numCols, 2, i + 1)

AttributeError: 'Model' object has no attribute 'x'

Changing it to the following works:
ace.plot_transforms(myace.ace, fname = 'mytransforms.pdf')

I'm not sure if this is an error on my part or from something else, but I cannot find the package "mace" on PyPI.

This script doesn't appear to run ever though, so I'm not sure it is too important.

The random seed used in many of the files is too large when using windows 64-bit.

    numpy.random.seed(9287349087)
  File "mtrand.pyx", line 646, in mtrand.RandomState.seed (numpy\random\mtrand\mtrand.c:7697)
ValueError: Seed must be between 0 and 4294967295

I am not sure if this is an issue, or a request for improvement/documentation.

In your ACE implementation, is there a way to expose directly the equations/forms for the transformations? If not, what would you recommend, running OLS in pairs (e.g. phi0-x0, phi1-x1, and so on)?

Hi there.
I am able to run the Wang example without problems.
However, when I try to adapt to use with my own data I get an error message, attached below:

Do you know what might be causing this? I am including my data to allow you to try and replicate the issue. I realize it is not a lot of data points but I suspect this has not to do with the issue.

data_final.txt

Follow up on a closed issue I have more questions.
I went back and reviewed the literature papers; if I understand those examples correctly, I think that one could use ACE in a predictive sense in a couple of ways:

1. use the magnitude of the transforms as a measure of the strength of the relationship between the original independent predictors and the target, even though as you say, there are no functional forms for the transforms

2. predict the target given new measurements of the predictors, using the inverse relationship between theta and Y; e.g., using the example from Wang and Murphy:
   
   or in pictorial way using my example from here:
   

Originally posted by @mycarta in #11 (comment)

The excellent test problem in [Wang, "Estimating Optimal Transformations for Multiple Regression Using the ACE Algorithm"] is acting up a little. On one hand, the basic shapes of the various components are being recovered. On the other hand, they are very noisy and not nearly as good as in the paper. This suggests something is slightly wrong in ACE still.

The initial implementation of ACE that's active does not make use of the update capability of the fixed-span smoother. This makes ACE really slow, as confirmed by this profiling result (made with gprof2dot):

Next step is to just modify the fixed-span smoother to update intelligently. Should be easy.

Maximal Correlation (MC) values should always be between 0 and 1. However, when I calculate the MC values of x1 and x2 with y for values of
x1 = [ 1., 2., 3., 4., 5., 6., 7., 8., 9., 10.]
x2 = [ 2., 5., 9., 7., 4., 8., 1., 6., 3., 10.]
y = [ 3., 9., 11., 8., 4., 15., 14., 20., 30., 32.]
I get a negative MC between x2 and y.

Running the same problem using the R library acepack yields an MC value within the proper range.

Python calculation:

def ACE(x, y):
    ''' 
    Output MCs: Maximal Correlations (MCs) for each variable x 
    Input x: list of 1D numpy arrays, one for each input variable
    Input y: 1D numpy array of responses
    '''
    ace_solver = ace.ACESolver()
    ace_solver.specify_data_set(x, y)
    ace_solver.solve()
    MCs = [] # mutual correlations
    for i in range(len(x)):
        (MC, Pval) = stats.pearsonr( ace_solver.x_transforms[i], ace_solver.y_transform )
        MCs.append( MC )
    return(MCs)

from ace import ace
from scipy import stats
import numpy as np
x = [np.array([ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  9., 10.]), np.array([ 2.,  5.,  9.,  7.,  4.,  8.,  1.,  6.,  3., 10.])]
y = np.array([ 3.,  9., 11.,  8.,  4., 15., 14., 20., 30., 32.])
MCs = ACE(x, y)
print('MCs = ', MCs)

yields
MCs = [0.9523, -0.0577]

Meaning the Maximal Correlation value between x2 and y is -0.058.

R acepack calculation:

library(acepack)
x1 = 1:10
x2 = c(2.,  5.,  9.,  7.,  4.,  8.,  1.,  6.,  3., 10.)
x <- cbind(x1, x2)
y = c( 3.,  9., 11.,  8.,  4., 15., 14., 20., 30., 32.)
ace_model = ace(x, y)
MC = cor(ace_model$tx, ace_model$ty)

yields MC values of

x1 0.9427068
x2 0.3442552

Giving a positive Maximal Correlation value between x2 and y of 0.344

The ACE algorithm is running and converging, but it doesn't converge to the right answer. What is the issue?

There's a good potential that I'm not applying the expected conditionals correctly. Right now I'm just using the supersmoother S(y|x) as a direct replacement for E(y|x). The FORTRAN code is too unreadable to see how it's supposed to be done.

I ran the sample wang example and ran to the issue because it does a
num_cols = len(ace_model.x) / 2 + 1 equal to 3.5.
that cause the value of num_cols be 3.5. I removed one of arrays in the x. Now I have 4 x samples. But again plotting threw the error because it is considering it as float .

ValueError: Number of rows must be a positive integer, not 3.0

PS.: "Successfully installed ace-0.3.2"