module__n_inputs=X.shape[-1],
module__n_hiddens=10,
module__n_layers=1,
optimizer=torch.optim.SGD,
optimizer__lr=1.00,
)
# generate some synthetic data
# t = 1000; d = 5; n = 100; X = (torch.rand([n,t,d])).double(); y = (torch.randint(2,(n,1))).reshape(-1).long()
rnn.fit(X, y)
### Evaluate
pd.set_option('display.precision', 3)
for n in range(dataset.N):
X, y = dataset.get_single_sequence_data(n)
yproba = rnn.forward(X)
result_df = pd.DataFrame(y, columns=['y_true'])
result_df['Pr(y=1)'] = yproba[:, 1]
print("\n### Seq %d" % n)
print(result_df.__str__())
"""
# Demonstrate the use of grid search by testing different learning
# rates while saving the best model at the end.
params = [
{
'lr': [10,20,30],
},
]
module__n_layers=1,
optimizer=torch.optim.SGD,
optimizer__lr=.01,
)
X, y = dataset.get_batch_data(batch_id=0)
rnn.fit(X, y)
### Evaluate
pd.set_option('display.precision', 3)
proba_0 = []
proba_1 = []
for n in range(dataset.N):
#print(n)
X, y = dataset.get_single_sequence_data(n)
yproba = float(rnn.forward(X)[:, 1])
if y[0] == 1:
proba_1.append(yproba)
else:
proba_0.append(yproba)
#result_df = pd.DataFrame(y, columns=['y_true'])
#result_df['Pr(y=1)'] = yproba[:,1]
# print("\n### Seq %d" % n)
# print(result_df.__str__())
print(np.mean(proba_0), np.mean(proba_1))
"""
# Demonstrate the use of grid search by testing different learning
# rates while saving the best model at the end.
params = [