def cosim_search(e,q=None,qw=None):
if (q is None) or (qw is None):
return cosim(e,e)
else:
idx = [i for i,v in enumerate(qw) if v != 0]
e = e.as_matrix()
e_idx = e[:,idx]
q_idx = q[idx].reshape(1,-1)
return cosim(q_idx,e_idx).flatten()
def cosim_search(e, q, qw):
idx = [i for i, v in enumerate(qw) if v != 0]
e = e.as_matrix()
e_idx = e[:, idx]
q_idx = q[idx].reshape(1, -1)
return cosim(q_idx, e_idx).flatten()
def calc_cosine_similarity(data_matrix):
# Calculate cosine similarity matrix
cosim_matrix = cosim(data_matrix)
# Calculate upper half of the cosim matrix without diagonal
upper_triangle = cosim_matrix[np.triu_indices_from(cosim_matrix, k=1)]
# Find matrix minimum and maximum greater than zero
print("Cosim Matrix Minimum: " + str(np.amin(upper_triangle)))
print("Cosim Matrix Maximum: " + str(np.amax(upper_triangle)))
return cosim_matrix
def response(user_response):
robo_response = ''
ChatResponse.sent_token.append(user_response)
TfidfVec = tfvec(tokenizer=lemitizer.LemNormalize,
stop_words='english')
tfidf = TfidfVec.fit_transform(ChatResponse.sent_token)
vals = cosim(tfidf[-1], tfidf)
idx = vals.argsort()[0][-2]
flat = vals.flatten()
flat.sort()
req_tfidf = flat[-2]
if (req_tfidf == 0):
robo_response = robo_response + "I am sorry! I don't understand you"
return robo_response
else:
robo_response = robo_response + ChatResponse.sent_token[idx]
return robo_response
def cosine_sim(x):
li = []
for item in x["sent_emb"][0]:
li.append(cosim(item,x["quest_emb"][0][0]))
return li
from sklearn.metrics.pairwise import cosine_similarity as cosim
compSim = np.zeros([np.size(compMat, 0), np.size(compMat, 0)])
for i in range(np.size(compMat, 0)):
print(i)
for j in range(np.size(compMat, 0)):
a1 = np.asarray(compBool[i, :]).astype(np.bool)
a2 = np.asarray(compBool[j, :]).astype(np.bool)
if (a1.sum() + a2.sum() == 0):
compSim[i, j] = 1
else:
intersection = np.logical_and(a1, a2)
compSim[i, j] = 2. * intersection.sum() / (a1.sum() + a2.sum())
print(compSim[1, i])
kinaseSim = cosim(kinaseFeat)
nt = len(kinaseSim)
#%% Clustering
import matplotlib.pyplot as plt
from scipy.spatial.distance import squareform
from scipy.spatial.distance import pdist
d = pdist(compMat, metric='hamming')
from scipy.cluster.hierarchy import linkage
z = linkage(d, method='complete')
from scipy.cluster.hierarchy import dendrogram
plt.figure()
dn = dendrogram(z)
import embed_functions as emb
from glob import glob
def cosim_search(e,q=None,qw=None):
if (q is None) or (qw is None):
return cosim(e,e)
else:
idx = [i for i,v in enumerate(qw) if v != 0]
e = e.as_matrix()
e_idx = e[:,idx]
q_idx = q[idx].reshape(1,-1)
return cosim(q_idx,e_idx).flatten()
path_work = '/mnt/HA/groups/rosenGrp/embed_cons/out/kegg'
params = ['6_256_5_50_10_1e-06_100','10_256_5_50_10_1e-06_100']
suffix = ['remb.csv.gz','remb_raw.csv.gz']
for p in params:
path_emb = glob(path_work + '/*' + p + '*.csv.gz')
for suf in suffix:
mods = [path for path in path_emb if suf in path]
if 'vcons' not in mods[0]:
mods = [mods[1],mods[0]]
remb_cons = pd.read_csv(mods[0],index_col=0)
remb_clust = pd.read_csv(mods[1],index_col=0)
sim = pd.DataFrame(cosim(remb_cons.as_matrix(),remb_clust.as_matrix()),
index=remb_cons.index,columns=remb_clust.index)
sim.to_csv(os.path.join(path_work,'cosim_cons_' + p + '_1e-05_' + suf))