Create custom Loss of MSE loss (weighted MSE loss) and L1 penality (L1 regulizer).
weight - of size input, weight of class for each value
"""
if isinstance(weighted, torch.Tensor):
return torch.sum(weighted * (input - target)**2)
else:
return crit(input, target)
# Get data
backup_dir = "backup_cpkl/"
med_name = "PsychroMesoThermHyperMedium"
#Get the scope matrix by running scope_kegg_prk.py
simplified_matrix = pk.load(open(
joinP(backup_dir, "simplified_matrix_scope{}_prk.cpk".format(med_name)),
"rb"),
encoding='latin1') #Scope matrix
df_species = pd.read_csv("species_metadata.csv", index_col=0)
df_species.loc[df_species.sp_codes.isnull(),
"sp_codes"] = "nan" # Otherwise interpreted as NaN
# Standardisation
growth_temp_crT = (df_species.temp_def.dropna().values -
np.min(df_species.temp_def.dropna().values)) / (
np.max(df_species.temp_def.dropna().values) -
np.min(df_species.temp_def.dropna().values))
matrix_tempT = simplified_matrix[:, df_species.temp_def.notnull()]
classes_T = df_species.temp_range_deduced[
df_species.temp_range_deduced.notnull()].values
valid_species_index = []
for i, code in enumerate(df_species.sp_codes):
old_dir = None
print "\n", i, df_species.iloc[i,[0, 1]].values
try :
obj = MetabolicGraph(df_species.iloc[i, 1], "", #no need for fasta for brite reconstruction
code=code, work_dir=org_dir,
through_brite=through_bri)
except SystemExit, IOError:
continue
try:
obj.directory = joinP(org_dir,
df_species.iloc[i, 1].replace("/", ".").replace(" ", "_")
+ "_" + df_species.iloc[i, 0])
obj.get_reaction_graph(gname="metabolites_reaction_" + code + ".graphml",
pklname="metabolites_reactions_graph_" + code + ".cpkl",
dir_ec="EC_global/", rn_dir="reaction_files/",
through_brite=through_bri)
obj.build_reaction_graph(filtr=False,
gname="metabolites_reaction_" + code + ".graphml",
pklname="metabolites_reactions_graph_" + code + ".cpkl")
valid_species_index.append(i)
except (SystemExit, IOError, TypeError): #Problem when constructing
os.system("rm -rf " + obj.directory) #remove directory
logger.error("Species %s %s will not be handled" %(df_species.iloc[i, 1], code))
continue
work_dir = "graph_species/" # working directory : directory where we can find our graphs
backup_dir = "backup_cpkl/"
if not os.path.exists(work_dir):
logger.error("Directory with graphs %s could not be found" % work_dir)
raise SystemExit()
if not os.path.exists(backup_dir):
logger.error("Backup directory backup_cpkl/ could not be found")
raise SystemExit()
df_species = pd.read_csv("species_metadata.csv", index_col=0)
df_species.loc[df_species.sp_codes.isnull(),
"sp_codes"] = "nan" # Otherwise interpreted as NaN
#either get backup or generate scope
if os.path.exists(joinP(backup_dir, "simplified_matrix_scope{}_prk.cpk".format(med_name))) and\
os.path.exists(joinP(backup_dir, "simplified_nodes_scope{}_prk.cpk".format(med_name))):
simplified_matrix = cpk.load(
open(
joinP(backup_dir,
"simplified_matrix_scope{}_prk.cpk".format(med_name)),
"rb"))
nodes_simplified = cpk.load(
open(
joinP(backup_dir,
"simplified_nodes_scope{}_prk.cpk".format(med_name)),
"rb"))
else:
all_scope = []
all_all_nodes = []
"#a15d90", "#a7bc74", "#5c6a9f", "#a65d22", "#8a4768", "#d6ac66", "#ee90a1", "#758348", "#c7726c",
"#746015", "#e89871", "#8e651e", "#944d32", "#896f3b", "#b78050"]
colours3 = ["blue", "#f32a30", "#de81d3", "#98cd28", "#278300","#fb2376", "#6adc90","#45124e", "#0082ad",
"#7c4f00", "#5a8fff","#dbc672", "#bd9eff","#ffa988", "#535254","#fdb0cd", "#b20085", "#54d3ff",] #red b90a06, red i like #fc4d45 violeta 8a335b, blue 0049c3
backup_dir = "backup_cpkl/"
med_name = "PsychroMesoThermHyperMedium"
#Get species metadata
df_species = pd.read_csv("species_metadata.csv", index_col=0)
df_species.loc[df_species.sp_codes.isnull(), "sp_codes"] = "nan" # Otherwise interpreted as NaN
#Get the scope matrix by running scope_kegg_prk.py :
simplified_matrix = pk.load(open(joinP(backup_dir, "simplified_matrix_scope{}_prk.cpk".format(med_name)), "rb"), encoding='latin1')
plotNull = True #Whether to plot species without metadata or not
perp = 40 #t-SNE perplexity
distance_tsne = "jaccard" #Distance to use for t-SNE
#Evaluating distance matrix
if distance_tsne == "jaccard":
if exists("backup_cpkl/jaccard_simi.pk"):
similarity = pk.load(open("backup_cpkl/jaccard_simi.pk", "rb"))
else:
similarity = sim_matrix(simplified_matrix.T) #takes some time
dist = 1 - similarity
elif distance_tsne == "manhattan":
dist = manhattan_distances(simplified_matrix.T)
])
break
if plot:
plt.ylabel("Accuracy")
plt.xlabel("Position of first compound in mask")
plt.title("Mask size: {}".format(step))
plt.plot()
return accs
if __name__ == '__main__':
backup_dir = "backup_cpkl/"
med_name = "PsychroMesoThermHyperMedium"
#The following files are generated by scope_kegg_prk.py
simplified_matrix = cpk.load(open(joinP(backup_dir, "simplified_matrix_scope{}_prk.cpk".format(med_name)), "rb"))
nodes_simplified = cpk.load(open(joinP(backup_dir, "simplified_nodes_scope{}_prk.cpk".format(med_name)), "rb"))
df_species = pd.read_csv("species_metadata.csv", index_col=0)
df_species.loc[df_species.sp_codes.isnull(), "sp_codes"] = "nan" # Otherwise interpreted as NaN
# =============================================================================
#
# TEMPERATURE CLASS PREDICTION
#
# =============================================================================
matrix_temp = simplified_matrix[:, df_species.temp_range_deduced.notnull()]