def update_model(self,sequences, bootstrap=True):
X=[]
Y=[]
for sequence in sequences:
if sequence not in self.one_hot_sequences:# or self.batch_update:
x=translate_string_to_one_hot(sequence,self.alphabet)#.flatten()
y=self.measured_sequences[sequence]
self.one_hot_sequences[sequence]=(x,y)
X.append(x)
Y.append(y)
else:
x,y=self.one_hot_sequences[sequence]
X.append(x)
Y.append(y)
X=np.array(X)
Y=np.array(Y)
for i,model_dict in enumerate(self.list_of_models_dict):
try:
y_pred=model_dict['model'].predict(X)
# self.noise_alpha=explained_variance_score(Y,y_pred)
self.model_performances[i]=r2_score(Y,y_pred)
except:
pass
indices=[]
for k in range(int(len(X)/2)):
indices.append(random.randint(0,len(X)-1))
model_dict['model'].fit(np.take(X,indices,axis=0),np.take(Y,indices,axis=0),epochs=model_dict['epochs'],validation_split=0,batch_size=model_dict['batch_size'],verbose=0)
best_model_index=np.argmax(self.model_performances)
self.best_model=self.list_of_models_dict[best_model_index]['model']
def update_model(self,sequences):
X=[]
Y=[]
for sequence in sequences:
if sequence not in self.one_hot_sequences:# or self.batch_update:
x=translate_string_to_one_hot(sequence,self.alphabet)#.flatten()
y=self.measured_sequences[sequence]
self.one_hot_sequences[sequence]=(x,y)
X.append(x)
Y.append(y)
else:
x,y=self.one_hot_sequences[sequence]
X.append(x)
Y.append(y)
X=np.array(X)
Y=np.array(Y)
try:
y_pred=self.skmodel.predict(X)
# self.noise_alpha=explained_variance_score(Y,y_pred)
self.noise_alpha=r2_score(Y,y_pred)
except:
pass
self.skmodel.fit(X,Y,epochs=self.epochs,validation_split=self.validation_split,batch_size=self.batch_size,verbose=0)
if not self.batch_update:
self.retrain_model()
def update_model(self,sequences):
X=[]
Y=[]
for sequence in sequences:
if sequence not in self.one_hot_sequences:# or self.batch_update:
#print (sequence)
x=translate_string_to_one_hot(sequence,self.alphabet).flatten()
y=self.measured_sequences[sequence]
self.one_hot_sequences[sequence]=(x,y)
X.append(x)
Y.append(y)
else:
x,y=self.one_hot_sequences[sequence]
X.append(x)
Y.append(y)
X=np.array(X)
Y=np.array(Y)
self.retrain_model()
try:
y_pred=self.skmodel.predict(X)
self.noise_alpha=explained_variance_score(Y,y_pred)
self.noise_alpha=r2_score(Y,y_pred)
except:
pass
self.skmodel.fit(X,Y)
def bootstrap(self,wt,alphabet):
sequences=[wt]
self.wt=wt
self.alphabet=alphabet
for i in range(len(wt)):
tmp=list(wt)
for j in range(len(alphabet)):
tmp[i]=alphabet[j]
sequences.append("".join(tmp))
self.measure_true_landscape(sequences)
self.one_hot_sequences={sequence:(translate_string_to_one_hot(sequence,self.alphabet),self.measured_sequences[sequence]) for sequence in sequences} #removed flatten for nn
self.update_model(sequences)
def get_uncertainty(self,sequence):
x=np.array([translate_string_to_one_hot(sequence,self.alphabet)])#.flatten()]
x_predicts=[m["model"].predict(x)[0][0] for m in self.list_of_models_dict]
return np.mean(x_predicts), np.std(x_predicts), x_predicts
def _fitness_function(self,sequence):
x=np.array([translate_string_to_one_hot(sequence,self.alphabet)])#.flatten()]
return max(min(200, self.best_model.predict(x)[0][0]),-200)