def sneonly(self,reinit,directory):
print("Performing sneonly")
inity = "{}/inity.npy".format(directory)
if (reinit):
try:
os.remove(inity)
except OSError:
print "No initial inity file"
if (self.pointslen > 2000):
print "Too many proteins to perform t-SNE directly"
sys.exit(2)
matrix = tsne_calc.tsne(inity,False,scipymat.toarray(),
no_dims=self.dim,
initial_dims=self.pointslen)
return matrix
def sneonly(self, reinit, directory):
print("Performing sneonly")
inity = "{}/inity.npy".format(directory)
if (reinit):
try:
os.remove(inity)
except OSError:
print "No initial inity file"
if (self.pointslen > 2000):
print "Too many proteins to perform t-SNE directly"
sys.exit(2)
matrix = tsne_calc.tsne(inity,
False,
scipymat.toarray(),
no_dims=self.dim,
initial_dims=self.pointslen)
return matrix
def main():
## Get the starting time to measure the time of the run
print "Running script..."
t0 = time.clock()
## Obtain all the info from the arguments passed
args = readargs.arg_parser()
print "Parsed arguments"
"""
If necessary, run BLAST locally (try to run on a faster machine instead)
#initrun.run_blast_tab(queryname,dbname,outfile,fmt,dbsize,ecutoff)
"""
## Define the paths of BLAST results file and names file
resultsfile = args.directory + "/results.out"
fastafile = args.directory + "/" + args.directory + ".fas"
matrixpath = args.directory + "/temp/mds.hdf5"
coordspath = args.directory + "/temp/" + args.directory + "_" + args.type + "_" + "coords.npy"
jsonpath = args.directory + "/temp/file.json"
inity = args.directory + "/temp/inity.npy"
if (args.reinitialize):
try:
os.remove(inity)
except OSError:
pass
## Obtain colors and names from names file
names, colors, lines, seqs = initrun.read_fasta(fastafile, args.format)
print "Read FASTA file"
## Obtain the handle to the results file
tabParser, tabHandle = initrun.open_file(resultsfile)
print "Opened file"
## Check if coordinates have already been mapped
if (args.load):
print "Loading coordinates from path"
matrix = np.load(args.load)
elif (args.precoordinated == True):
print "Loading coordinates from temp"
matrix = np.load(coordspath)
else:
## Check if results are preparsed
##
## If no, create an HDF5 formatted matrix and initialize
## Then, parse BLAST results and populate the HDF5 matrix
##
## If yes, then obtain populated matrix from file
if args.preparsed:
print "Retrieving matrix"
hdfmat = initrun.get_matrix(matrixpath)
else:
hdfmat = initrun.create_matrix(args.value, names, matrixpath)
print "Initialized matrix"
print "Parsing results"
if (args.format == 'mod'):
results_parser.next_line_modified_format(
args.value, tabParser, tabHandle, names, hdfmat)
elif (args.format == 'orig'):
results_parser.next_line_original_format(
args.value, tabParser, tabHandle, names, hdfmat)
## Run the appropriate dimensionality reduction algorithm
## -mdsonly = metric MDS with sklearn's manifold package
## -snemds = preprocess to "points/10" dimensions with MDS, then t-SNE for reduced matrix
## -snepca = preprocess to "points/10" dimensions with PCA, then t-SNE for reduced matrix
## (still working on -n = nystrom MDS with pycogent's approximate_mds package)
if (args.type == "mdsonly"):
print "Performing MDS"
matrix = mds_calc.metric_mds(hdfmat, int(args.dimension))
elif (args.type == "snemds"):
print "Performing t-SNE with MDS preprocessing"
## Partially reduce dimensionality of HDF5 matrix to 1/10th of original size or maximum of 400
tempred = min(int(len(names) / 10), 400)
print "Preprocessing the data using MDS..."
print "Reducing to", tempred, "dimensions"
tempmatrix = mds_calc.metric_mds(hdfmat, tempred)
matrix = tsne_calc.tsne(inity,
False,
tempmatrix,
no_dims=int(args.dimension),
initial_dims=tempred)
elif (args.type == "snepca"):
print "Performing t-SNE with PCA preprocessing"
## Partially reduce dimensionality of HDF5 matrix to 1/10th of original size or maximum of 400
tempred = min(int(len(names) / 10), 400)
print "Reducing to", tempred, "dimensions"
matrix = tsne_calc.tsne(inity,
True,
hdfmat,
no_dims=int(args.dimension),
initial_dims=tempred)
elif (args.type == "sneonly"):
print "CAUTION: Performing t-SNE on full dissimilarity matrix"
if (len(names) > 2000):
print "Too many proteins to perform t-SNE directly"
sys.exit(2)
matrix = tsne_calc.tsne(inity,
False,
hdfmat[...],
no_dims=int(args.dimension),
initial_dims=len(names))
# model = TSNE(n_components=3,metric="precomputed")
# matrix = model.fit_transform(hdfmat)
#else: matrix = mds_calc.nystrom_frontend(len(names),math.sqrt(len(names)),2,mds_calc.getdist,hdfmat)
# save coordinates to file
np.save(coordspath, matrix)
print "Took", time.clock() - t0, "seconds"
#with open(jsonpath, 'w') as jsonout:
# json.dump(jsonconv.jsonmaker(colors,lines,matrix,args.format), jsonout, indent=2)
## Plot the results with matplotlib's PyPlot
if (args.plot):
print "Plotting", len(names), "points"
if (int(args.dimension) == 2):
plotter.pyplotter2d(matrix, colors, names, seqs, args.directory)
elif (int(args.dimension) == 3):
plotter.pyplotter3d(matrix, colors)
def main():
## Get the starting time to measure the time of the run
print "Running script..."
t0 = time.clock()
## Obtain all the info from the arguments passed
args = readargs.arg_parser()
print "Parsed arguments"
"""
If necessary, run BLAST locally (try to run on a faster machine instead)
#initrun.run_blast_tab(queryname,dbname,outfile,fmt,dbsize,ecutoff)
"""
## Define the paths of BLAST results file and names file
resultsfile = args.directory + "/results.out"
fastafile = args.directory + "/" + args.directory + ".fas"
matrixpath = args.directory + "/temp/mds.hdf5"
coordspath = args.directory + "/temp/" + args.directory + "_"+ args.type + "_" + "coords.npy"
jsonpath = args.directory + "/temp/file.json"
inity = args.directory + "/temp/inity.npy"
if (args.reinitialize):
try:
os.remove(inity)
except OSError:
pass
## Obtain colors and names from names file
names,colors,lines,seqs = initrun.read_fasta(fastafile,args.format)
print "Read FASTA file"
## Obtain the handle to the results file
tabParser, tabHandle = initrun.open_file(resultsfile)
print "Opened file"
## Check if coordinates have already been mapped
if (args.load):
print "Loading coordinates from path"
matrix = np.load(args.load)
elif (args.precoordinated == True):
print "Loading coordinates from temp"
matrix = np.load(coordspath)
else:
## Check if results are preparsed
##
## If no, create an HDF5 formatted matrix and initialize
## Then, parse BLAST results and populate the HDF5 matrix
##
## If yes, then obtain populated matrix from file
if args.preparsed:
print "Retrieving matrix"
hdfmat = initrun.get_matrix(matrixpath)
else:
hdfmat = initrun.create_matrix(args.value,names,matrixpath)
print "Initialized matrix"
print "Parsing results"
if (args.format == 'mod'):
results_parser.next_line_modified_format(args.value,tabParser,tabHandle,names,hdfmat)
elif (args.format == 'orig'):
results_parser.next_line_original_format(args.value,tabParser,tabHandle,names,hdfmat)
## Run the appropriate dimensionality reduction algorithm
## -mdsonly = metric MDS with sklearn's manifold package
## -snemds = preprocess to "points/10" dimensions with MDS, then t-SNE for reduced matrix
## -snepca = preprocess to "points/10" dimensions with PCA, then t-SNE for reduced matrix
## (still working on -n = nystrom MDS with pycogent's approximate_mds package)
if (args.type == "mdsonly"):
print "Performing MDS"
matrix = mds_calc.metric_mds(hdfmat,int(args.dimension))
elif (args.type == "snemds"):
print "Performing t-SNE with MDS preprocessing"
## Partially reduce dimensionality of HDF5 matrix to 1/10th of original size or maximum of 400
tempred = min(int(len(names)/10),400)
print "Preprocessing the data using MDS..."
print "Reducing to", tempred, "dimensions"
tempmatrix = mds_calc.metric_mds(hdfmat,tempred)
matrix = tsne_calc.tsne(inity,False,tempmatrix,no_dims=int(args.dimension),initial_dims=tempred)
elif (args.type == "snepca"):
print "Performing t-SNE with PCA preprocessing"
## Partially reduce dimensionality of HDF5 matrix to 1/10th of original size or maximum of 400
tempred = min(int(len(names)/10),400)
print "Reducing to", tempred, "dimensions"
matrix = tsne_calc.tsne(inity,True,hdfmat,no_dims=int(args.dimension),initial_dims=tempred)
elif (args.type == "sneonly"):
print "CAUTION: Performing t-SNE on full dissimilarity matrix"
if (len(names) > 2000):
print "Too many proteins to perform t-SNE directly"
sys.exit(2)
matrix = tsne_calc.tsne(inity,False,hdfmat[...],no_dims=int(args.dimension),initial_dims=len(names))
# model = TSNE(n_components=3,metric="precomputed")
# matrix = model.fit_transform(hdfmat)
#else: matrix = mds_calc.nystrom_frontend(len(names),math.sqrt(len(names)),2,mds_calc.getdist,hdfmat)
# save coordinates to file
np.save(coordspath,matrix)
print "Took", time.clock()-t0, "seconds"
#with open(jsonpath, 'w') as jsonout:
# json.dump(jsonconv.jsonmaker(colors,lines,matrix,args.format), jsonout, indent=2)
## Plot the results with matplotlib's PyPlot
if (args.plot):
print "Plotting",len(names), "points"
if (int(args.dimension) == 2):
plotter.pyplotter2d(matrix,colors,names,seqs,args.directory)
elif (int(args.dimension) == 3):
plotter.pyplotter3d(matrix,colors)
