import os
import numpy
import pandas
from pandas import DataFrame
from sklearn.externals import joblib
from sklearn.decomposition import ProjectedGradientNMF
from sklearn.decomposition import PCA
from math import sqrt

# load raw_clean_dedupe_norm into memory
dodge_data = joblib.load(os.environ['DATASET_D'])

# initialize a PCA with arbitrary number of components
pca = PCA(n_components=dodge_data.shape[1])

# fit the pca to raw_clean_dedupe_norm
pca.fit(dodge_data)

# set the number of principle components equal to the first n components that explain 95% of the variance
explained_variance = pca.explained_variance_ratio_.cumsum()
explained_variance = explained_variance[explained_variance <= .95]
p_comp = len(explained_variance)

# use the number of principle components to initialize a matrix factorization model
model = ProjectedGradientNMF(n_components=p_comp,
                             init='nndsvd',
                             beta=1,
                             sparseness=None)

# fit NMF model to the raw_clean_dedupe_norm dataset
model.fit(dodge_data)

# load the NMF model that was used to compute the original kmeans centroids
nmf = joblib.load(os.environ['MODEL_A'])

# load the kmeans object
kmeans = joblib.load(os.environ['MODEL_B'])
kmeans = kmeans[6]

# transform the NMF centroids back into the original 35-column feature-space
centroids = numpy.dot(kmeans.cluster_centers_, nmf.components_)

# transform the 35-column centroids into the newly computed NMF matrix
centroids = model.transform(centroids)

def normalizeZeroOne(vec):
    normed = (vec - min(vec)) / (max(vec) - min(vec))
    return normed

#normalize the transformed centroid features
normed_centroids = numpy.zeros(centroids.shape)
for i in range(0, centroids.shape[1]):
    normed_centroids[:, i] = normalizeZeroOne(centroids[:, i])



normed_centroids = numpy.zeros(centroids.shape)
for i in range(0, centroids.shape[0]):
    normed_centroids[i, :] = normalizeZeroOne(centroids[i, :])


pandas.DataFrame(normed_centroids).to_csv('/users/zpuste/digital-personas/data/dodge/mobile/nmf_transformed_centroids.csv')

nmf_comps = pandas.DataFrame(model.components_)

nmf_comps.columns = dodge_data.columns
nmf_comps.to_csv('/users/zpuste/digital-personas/data/dodge/mobile/nmf_components.csv')