def nnmf(nfact, method="scd", loss="mse", period_min=60):
print "\n\n\n***Performing non-negative-matrix factorization with " + method + " method and " + loss + " loss function...\n"
ro.r('fit<-nnmf(x, ' + nfact + ', max.iter = 100000, rel.tol = -1, method = "' + method + '",inner.max.iter = 1, loss="' + loss + '", n.threads=20);')
print "\nPrinting a summary of the fitting matrix with method " + method + " and loss " + loss + "...\n"
ro.r('summary(fit)') # print variance accounted for
print("\n")
print "\nPrinting the fitting matrix with method " + method + " and loss " + loss + "...\n"
ro.r('print(fit)')
print("\n")
print "\nPrinting the score matrix with method " + method + " and loss " + loss + "...\n"
ro.r('print(fit$W, digits=2, cutoff=0.4)')
print("\n")
print "\nPrinting the factor matrix with method " + method + " and loss " + loss + "...\n"
ro.r('print(t(fit$H), digits=2, cutoff=0.4)')
print("\n")
print "\nPlotting convergence to epoch with method " + method + " and loss " + loss + "...\n"
ro.r('jpeg(file = "' + output_folder + "/" + method + "_" + loss + '_convergence.jpeg")')
ro.r('plot(NULL, xlim = c(1, 3000), ylim = c(0.15, 0.45), xlab = "Epochs", ylab = "' + loss + '");')
ro.r('lines(cumsum(fit$average.epochs), fit$' + loss + ');')
ro.r('dev.off()')
print("\n")
print "\nPrinting the loss and the cumsum with method " + method + " and loss " + loss + "...\n"
ro.r('print(fit$loss, digits=2, cutoff=0.4)')
ro.r('print(cumsum(fit$average.epochs), digits=2, cutoff=0.4)')
print("\n")
print "\nPlotting the heatmaps with method " + method + " and loss " + loss + "...\n"
ro.r('jpeg(file = "' + output_folder + "/" + method + "_" + loss + '_heatmap_W.jpeg")')
ro.r('heatmap(fit$W, Colv = NA, xlab = "samples", ylab = "factors", margins = c(2,2),'+\
'labRow = "", labCol = "", scale = "column", col = cm.colors(100));')
ro.r('dev.off()')
ro.r('jpeg(file = "' + output_folder + "/" + method + "_" + loss + '_heatmap_H.jpeg")')
ro.r('heatmap(fit$H, Rowv = NA, ylab = "variables", xlab = "factors", margins = c(2,2),'+\
'labRow = '', labCol = '', scale = "row", col = cm.colors(100));')
ro.r('dev.off()')
print("\n")
loadings = np.asarray(ro.r('t(fit$H)'))
print "Loadings are: " + str(loadings)
factor_names = np.asarray([method + str(i) for i in range(loadings.shape[1])])
print "Factor names are: " + str(factor_names)
services_names = np.asarray(ro.r('colnames(fit$H)'))
print "Service names are: " + str(services_names)
traffic_matrix = np.asarray(ro.r('x'))
traffic_snapshots = np.asarray(ro.r('rownames(x)'))
traffic_services = np.asarray(ro.r('colnames(x)'))
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + loss)
all_dict = {}
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min)
all_dict["signatures"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, normalized_y=True)
all_dict["norm_signatures"] = temp_dict
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + loss, threshold=0.4, max_factor_value=True)
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, threshold=0.4, max_factor_value=True)
all_dict["signatures_th_0.4"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, normalized_y=True, threshold=0.4, max_factor_value=True)
all_dict["norm_signatures_th_0.4"] = temp_dict
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + loss, threshold=0.6, max_factor_value=True)
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, threshold=0.6, max_factor_value=True)
all_dict["signatures_th_0.6"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, normalized_y=True, threshold=0.6, max_factor_value=True)
all_dict["norm_signatures_th_0.6"] = temp_dict
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + loss, threshold=0.8, max_factor_value=True)
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, threshold=0.8, max_factor_value=True)
all_dict["signatures_th_0.8"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + loss + "/signatures/", period_min, normalized_y=True, threshold=0.8, max_factor_value=True)
all_dict["norm_signatures_th_0.8"] = temp_dict
# Analyzing scores
scores = np.asarray(ro.r('fit$W'))
print "Scores are: " + str(scores)
population_names = np.asarray(ro.r('rownames(fit$W)'))
print "Population names are: " + str(population_names)
score_factor_names = np.asarray([method + str(i) for i in range(loadings.shape[1])])
print "Score factor names are: " + str(score_factor_names)
scores_df = analyze_scores(scores, population_names, score_factor_names, method + "_" + loss)
print "Plotting scores (" + method + " and loss " + loss + ")...\n"
plot_factor_scores(factor_names, scores_df, method + "_" + loss, False)
plot_factor_scores(factor_names, scores_df, method + "_" + loss, True)
temp_dict = utilities.plot_time_signatures(scores_df[factor_names].as_matrix(), scores_df["snapshot_name"], factor_names, label, output_folder, method + "_" + loss + "/scores/", period_min)
all_dict["scores"] = temp_dict
plot_data_by_factor(all_dict, factor_names, label, output_folder, method + "_" + loss + "/summary_plots", period_min)
#plot_data_by_factor(all_dict, factor_names, label, output_folder, method + "_" + loss + "/summary_plots", period_min)
# Analyzing loadings
print "Saving loadings to file (" + method + " and loss " + loss + ")...\n"
loadings_df = analyze_loadings(loadings, factor_names, services_names, method + "_" + loss)
return loadings_df, scores_df
def efa(snapshot_filename,
nfact,
output_folder,
rotation="none",
method="ml",
cor=True):
# print "Dimension of x matrix:\n%s" % str(ro.r('dim(x)'))
# print "Head of x matrix:\n%s" % str(ro.r('head(x)'))
# ro.r('ev <- eigen(cor(x))') # get eigenvalues
# ro.r('ap <- parallel(subject=nrow(x),var=ncol(x),rep=100,cent=.05)')
# ro.r('nS <- nScree(x=ev$values, aparallel=ap$eigen$qevpea)')
# ro.r('plotnScree(nS)')
# compute num factors
print "\n\n\n***Performing factorial analysis with " + method + " fitting method, " + rotation + " rotational method...\n"
ro.r('fit <- fa(r=x,nfactors=' + nfact +
',n.iter=1, min.err = 0.001, max.iter = 50, \
rotate="' + rotation + '", scores="regression", \
residuals=TRUE, SMC=TRUE, missing=FALSE,impute="median",\
warnings=TRUE, fm="' + method + '",\
alpha=.1,p=.05,oblique.scores=FALSE,use="pairwise", \
covar= ' + ("F" if cor else "T") + ', cor = "' +
("cor" if cor else "cov") + '")')
print "\nPrinting a summary of the fitting matrix with method " + method + " and rotation " + rotation + "...\n"
ro.r('summary(fit)') # print variance accounted for
print("\n")
print "\nPrinting the fitting matrix with method " + method + " and rotation " + rotation + "...\n"
ro.r('print(fit)')
print("\n")
print "\nPrinting the loadings matrix with method " + method + " and " + rotation + " rotation...\n"
ro.r('print(loadings(fit), digits=2, cutoff=0.4)')
print("\n")
loadings = np.asarray(ro.r('fit$loadings'))
factor_names = np.asarray(ro.r('colnames(fit$loadings)'))
population_names = np.asarray(ro.r('rownames(fit$loadings)'))
traffic_matrix = np.asarray(ro.r('x'))
traffic_base_stations = np.asarray(ro.r('rownames(x)'))
traffic_snapshots = np.asarray(ro.r('colnames(x)'))
full_scores_df = pd.DataFrame()
for c in CATEGORIES:
searched_category = ',' + c + ','
output_subfolders = output_folder + "/" + method + "_" + rotation + "/" + c + "/"
if not os.path.exists(os.path.dirname(output_subfolders)):
os.makedirs(os.path.dirname(output_subfolders))
found_elements = [(i, (v.split(",")[0]))
for i, v in enumerate(population_names)
if searched_category in v]
if len(found_elements) == 0:
print("*** PROBLEM HERE: no element for category " + c)
continue
indices, c_traffic_snapshots = zip(*found_elements)
indices = list(indices)
c_traffic_base_stations = list(traffic_base_stations)
c_traffic_matrix = traffic_matrix[indices]
#print("*** For category " + c + "\nFound elements: " + str(found_elements) + "\nIndices: " + str(indices) + "\nc_traffic_matrix: " + str(c_traffic_matrix) + "\nCreated output folder: " + output_folder)
utilities.generate_signatures(c_traffic_matrix, c_traffic_base_stations, \
traffic_snapshots, loadings, factor_names, label, output_folder, method + "_" + rotation + "/" + c + "/")
utilities.generate_signatures(c_traffic_matrix, c_traffic_base_stations, \
traffic_snapshots, loadings, factor_names, label, output_folder, method + "_" + rotation + "/" + c + "/", threshold=0.4)
utilities.generate_signatures(c_traffic_matrix, c_traffic_base_stations, \
traffic_snapshots, loadings, factor_names, label, output_folder, method + "_" + rotation + "/" + c + "/", threshold=0.6)
utilities.generate_signatures(c_traffic_matrix, c_traffic_base_stations, \
traffic_snapshots, loadings, factor_names, label, output_folder, method + "_" + rotation + "/" + c + "/", threshold=0.8)
# Analyzing scores
scores = np.asarray(ro.r('fit$scores'))
population_names = np.asarray(ro.r('rownames(fit$scores)'))
score_factor_names = np.asarray(ro.r('colnames(fit$scores)'))
found_elements = [(i, (v.split(",")[0]))
for i, v in enumerate(population_names)
if searched_category in v]
if len(found_elements) == 0:
print("*** PROBLEM HERE: no element for category " + c)
continue
indices, c_traffic_snapshots = zip(*found_elements)
indices = list(indices)
c_population_names = list(c_traffic_snapshots)
c_scores = scores[indices]
scores_df = analyze_scores(c_scores, c_population_names,
score_factor_names, output_folder,
method + "_" + rotation + "/" + c + "/")
scores_df["category"] = c
full_scores_df = full_scores_df.append(scores_df, ignore_index=True)
output_filename = output_folder + "/" + method + "_" + rotation + "/all_scores" + (
"_" + label if label is not None else "") + ".csv"
full_scores_df.to_csv(output_filename, header=True, sep=",", index=False)
traffic_matrix = np.asarray(ro.r('x'))
traffic_base_stations = np.asarray(ro.r('rownames(x)'))
print "*** Here follows the base stations:"
print traffic_base_stations
traffic_snapshots = np.asarray(ro.r('colnames(x)'))
print "*** Here follows the traffic_snapshots:"
print traffic_snapshots
print "*** Here follows the traffic matrix:"
print traffic_matrix
# Analyzing loadings
print "Saving loadings to file (" + method + " and rotation " + rotation + ")...\n"
loadings_df = analyze_loadings(loadings, factor_names, traffic_snapshots,
output_folder, method + "_" + rotation)
print "Plotting loadings (" + method + " and rotation " + rotation + ")...\n"
plot_factor_loadings(factor_names, loadings_df, output_folder,
method + "_" + rotation, False)
plot_factor_loadings(factor_names, loadings_df, output_folder,
method + "_" + rotation, True)
# Analyzing uniqueness
print "\nPrinting the uniqueness matrix with method " + method + " and " + rotation + " rotation...\n"
ro.r('print(fit$uniquenesses)')
print("\n")
uniquenesses = np.asarray(ro.r('fit$uniquenesses'))
uniquenesses_names = np.asarray(ro.r('names(fit$uniquenesses)'))
print "Saving uniquenesses to file (" + method + " and rotation " + rotation + ")...\n"
uniquenesses_df = analyze_uniquenesses(uniquenesses, uniquenesses_names,
output_folder,
method + "_" + rotation)
print "Plotting uniquenesses (" + method + " and rotation " + rotation + ")...\n"
plot_factor_loadings(["uniquenesses"], uniquenesses_df, output_folder,
method + "_" + rotation, False, True)
return loadings_df, full_scores_df, uniquenesses_df
def efa(nfact, rotation="none", method="ml", cor=True, period_min=60):
# print "Dimension of x matrix:\n%s" % str(ro.r('dim(x)'))
# print "Head of x matrix:\n%s" % str(ro.r('head(x)'))
# ro.r('ev <- eigen(cor(x))') # get eigenvalues
# ro.r('ap <- parallel(subject=nrow(x),var=ncol(x),rep=100,cent=.05)')
# ro.r('nS <- nScree(x=ev$values, aparallel=ap$eigen$qevpea)')
# ro.r('plotnScree(nS)')
# compute num factors
print "\n\n\n***Performing factorial analysis with " + method + " fitting method, " + rotation + " rotational method...\n"
ro.r('fit <- fa(r=x,nfactors=' + nfact + ',n.iter=1, min.err = 0.001, max.iter = 50, \
rotate="' + rotation + '", scores="regression", \
residuals=TRUE, SMC=TRUE, missing=FALSE,impute="median",\
warnings=TRUE, fm="' + method + '",\
alpha=.1,p=.05,oblique.scores=FALSE,use="pairwise", \
covar= ' + ("F" if cor else "T") + ', cor = "' + ("cor" if cor else "cov") + '")')
print "\nPrinting a summary of the fitting matrix with method " + method + " and rotation " + rotation + "...\n"
ro.r('summary(fit)') # print variance accounted for
print("\n")
print "\nPrinting the fitting matrix with method " + method + " and rotation " + rotation + "...\n"
ro.r('print(fit)')
print("\n")
print "\nPrinting the loadings matrix with method " + method + " and " + rotation + " rotation...\n"
ro.r('print(loadings(fit), digits=2, cutoff=0.4)')
print("\n")
loadings = np.asarray(ro.r('fit$loadings'))
factor_names = np.asarray(ro.r('colnames(fit$loadings)'))
services_names = np.asarray(ro.r('rownames(fit$loadings)'))
traffic_matrix = np.asarray(ro.r('x'))
traffic_snapshots = np.asarray(ro.r('rownames(x)'))
traffic_services = np.asarray(ro.r('colnames(x)'))
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + rotation)
all_dict = {}
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min)
all_dict["signatures"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, normalized_y=True)
all_dict["norm_signatures"] = temp_dict
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + rotation, threshold=0.4)
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, threshold=0.4)
all_dict["signatures_th_0.4"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, normalized_y=True, threshold=0.4)
all_dict["norm_signatures_th_0.4"] = temp_dict
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + rotation, threshold=0.6)
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, threshold=0.6)
all_dict["signatures_th_0.6"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, normalized_y=True, threshold=0.6)
all_dict["norm_signatures_th_0.6"] = temp_dict
signatures, signatures_df, _ = utilities.generate_signatures(traffic_matrix, traffic_snapshots, \
traffic_services, loadings, factor_names, label, output_folder, method + "_" + rotation, threshold=0.8)
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, threshold=0.8)
all_dict["signatures_th_0.8"] = temp_dict
temp_dict = utilities.plot_time_signatures(signatures, signatures_df["index"], factor_names, label, output_folder, method + "_" + rotation + "/signatures/", period_min, normalized_y=True, threshold=0.8)
all_dict["norm_signatures_th_0.8"] = temp_dict
# Analyzing scores
scores = np.asarray(ro.r('fit$scores'))
population_names = np.asarray(ro.r('rownames(fit$scores)'))
score_factor_names = np.asarray(ro.r('colnames(fit$scores)'))
scores_df = analyze_scores(scores, population_names, score_factor_names, method + "_" + rotation)
print "Plotting scores (" + method + " and rotation " + rotation + ")...\n"
plot_factor_scores(factor_names, scores_df, method + "_" + rotation, False)
plot_factor_scores(factor_names, scores_df, method + "_" + rotation, True)
temp_dict = utilities.plot_time_signatures(scores_df[factor_names].as_matrix(), scores_df["snapshot_name"], factor_names, label, output_folder, method + "_" + rotation + "/scores/", period_min)
all_dict["scores"] = temp_dict
plot_data_by_factor(all_dict, factor_names, label, output_folder, method + "_" + rotation + "/summary_plots", period_min)
# Analyzing loadings
print "Saving loadings to file (" + method + " and rotation " + rotation + ")...\n"
loadings_df = analyze_loadings(loadings, factor_names, services_names, method + "_" + rotation)
print "Plotting loadings (" + method + " and rotation " + rotation + ")...\n"
plot_factor_loadings(factor_names, loadings_df, method + "_" + rotation, False)
plot_factor_loadings(factor_names, loadings_df, method + "_" + rotation, True)
# Analyzing uniqueness
print "\nPrinting the uniqueness matrix with method " + method + " and " + rotation + " rotation...\n"
ro.r('print(fit$uniquenesses)')
print("\n")
uniquenesses = np.asarray(ro.r('fit$uniquenesses'))
uniquenesses_names = np.asarray(ro.r('names(fit$uniquenesses)'))
print "Saving uniquenesses to file (" + method + " and rotation " + rotation + ")...\n"
uniquenesses_df = analyze_uniquenesses(uniquenesses, uniquenesses_names, method + "_" + rotation)
print "Plotting uniquenesses (" + method + " and rotation " + rotation + ")...\n"
plot_factor_loadings(["uniquenesses"], uniquenesses_df, method + "_" + rotation, False, True)
return loadings_df, scores_df, uniquenesses_df
