m = model.predict(x_test) #returns the median prediction if more than one tree
#evaluate results
mse = sklearn.metrics.mean_squared_error(y_test, m)
mabs = sklearn.metrics.mean_absolute_error(y_test, m)
exvar = sklearn.metrics.explained_variance_score(y_test, m)
print('MSE', mse)
print('M Abs Err', mabs)
print('Expl. Var.', exvar)
#close model
model.close_model()
print("Reload model and continue training for 20 epochs")
# reload model; can also open it using cPickle.load()
model2 = djinn.load(model_name="reg_djinn_test")
#continue training for 20 epochs using same learning rate, etc as before
model2.continue_training(x_train,
y_train,
20,
learnrate,
batchsize,
random_state=1)
#make updated predictions
m2 = model2.predict(x_test)
#evaluate results
mse2 = sklearn.metrics.mean_squared_error(y_test, m2)
mabs2 = sklearn.metrics.mean_absolute_error(y_test, m2)
def model_train1():
print("djinn example")
modelname = "class_djinn_test" # name the model
ntrees = 1 # number of trees = number of neural nets in ensemble
maxdepth = 4 # max depth of tree -- optimize this for each data set
dropout_keep = 1.0
model = djinn.DJINN_Classifier(ntrees, maxdepth, dropout_keep)
# find optimal settings: this function returns dict with hyper-parameters
# each djinn function accepts random seeds for reproducible behavior
optimal = model.get_hyperparameters(x_train, y_train, random_state=1)
batchsize = optimal['batch_size']
learnrate = optimal['learn_rate']
epochs = optimal['epochs']
# train the model with hyperparameters determined above
model.train(x_train,
y_train,
epochs=epochs,
learn_rate=learnrate,
batch_size=batchsize,
display_step=1,
save_files=True,
file_name=modelname,
save_model=True,
model_name=modelname,
random_state=1)
# *note there is a function model.fit(x_train,y_train, ... ) that wraps
# get_hyperparameters() and train(), so that you do not have to manually
# pass hyperparameters to train(). However, get_hyperparameters() can
# be expensive, so I recommend running it once per dataset and using those
# hyperparameter values in train() to save computational time
# make predictions
m = model.predict(
x_test) #returns the median prediction if more than one tree
model2 = djinn.load(model_name="class_djinn_test")
#evaluate results
'''with cf.ProcessPoolExecutor() as p:
acc=p.submit(sklearn.metrics.accuracy_score,y_test,m.flatten())
p.submit(model2.continue_training,x_train, y_train, 20, learnrate, batchsize, 1)
acc=acc.result()'''
acc = sklearn.metrics.accuracy_score(y_test, m.flatten())
print('Accuracy', acc)
#acc=sklearn.metrics.accuracy_score(y_test,m.flatten())
model2.continue_training(x_train,
y_train,
20,
learnrate,
batchsize,
random_state=1)
#close model
model.close_model()
print("Reload model and continue training")
# reload model; can also open it using cPickle.load()
#model2=djinn.load(model_name="class_djinn_test")
#continue training for 20 epochs using same learning rate, etc as before
#make updated predictions
m2 = model2.predict(x_test)
#evaluate results
acc = sklearn.metrics.accuracy_score(y_test, m.flatten())
print('Ye model ka Accuracy', acc)
# plt.show()
X=X.reshape(-1,1)
x_train,x_test,y_train,y_test=train_test_split(X, y, test_size=0.2, random_state=1)
print("djinn example")
modelname="gaussian_djinn_test" # name the model
ntrees=1 # number of trees = number of neural nets in ensemble
maxdepth=4 # max depth of tree -- optimize this for each data set
dropout_keep=1.0 # dropout typically set to 1 for non-Bayesian models
load_model = True
if load_model:
model=djinn.load(model_name=modelname)
else:
#initialize the model
model=djinn.DJINN_Regressor(ntrees,maxdepth,dropout_keep)
# find optimal settings: this function returns dict with hyper-parameters
# each djinn function accepts random seeds for reproducible behavior
optimal=model.get_hyperparameters(x_train, y_train, random_state=1)
batchsize=optimal['batch_size']
learnrate=optimal['learn_rate']
epochs=optimal['epochs']
# train the model with hyperparameters determined above
model.train(x_train,y_train,epochs=epochs,learn_rate=learnrate, batch_size=batchsize,
display_step=1, save_files=True, file_name=modelname,
save_model=True,model_name=modelname, random_state=1)
save_files=True,
file_name=modelname,
save_model=True,
model_name=modelname,
random_state=1)
# *note there is a function model.fit(x_train,y_train, ... ) that wraps
# get_hyperparameters() and train(), so that you do not have to manually
# pass hyperparameters to train(). However, get_hyperparameters() can
# be expensive, so I recommend running it once per dataset and using those
# hyperparameter values in train() to save computational time
# make predictions
m = model.predict(x_test) #returns the median prediction if more than one tree
model2 = djinn.load(model_name="class_djinn_test")
#evaluate results
with cf.ProcessPoolExecutor() as p:
acc = p.submit(sklearn.metrics.accuracy_score, y_test, m.flatten())
p.submit(model2.continue_training, x_train, y_train, 20, learnrate,
batchsize, 1)
acc = acc.result()
#acc=sklearn.metrics.accuracy_score(y_test,m.flatten())
model2.continue_training(x_train,
y_train,
20,
learnrate,
batchsize,
random_state=1)
print('Accuracy', acc)