def test_add_output_layer2(self):
# check if it reshapes the output correctly
model = Model(model={'layers': {
'lstm': 64,
'Dense': 1
}},
data=load_nasdaq(),
verbosity=0)
self.assertEqual(model._model.outputs[0].shape[1], model.outs)
self.assertEqual(model._model.outputs[0].shape[-1], model.forecast_len)
return
def test_add_output_layer1(self):
# check if it adds both dense and reshapes it correctly or not
model = Model(model={'layers': {
'lstm': 64
}},
data=load_nasdaq(),
verbosity=0)
self.assertEqual(model._model.outputs[0].shape[1], model.outs)
self.assertEqual(model._model.outputs[0].shape[-1], model.forecast_len)
return
def test_add_no_output_layer(self):
# check if it does not add layers when it does not have to
model = Model(model={
'layers': {
'lstm': 64,
'Dense': 1,
'Reshape': {
'target_shape': (1, 1)
}
}
},
data=load_nasdaq(),
verbosity=0)
self.assertEqual(model._model.outputs[0].shape[1], model.outs)
self.assertEqual(model._model.outputs[0].shape[-1], model.forecast_len)
return
def make_and_run(input_model, _layers=None, lookback=12, epochs=1, **kwargs):
df = load_nasdaq()
model = input_model(data=df,
verbosity=0,
batch_size=64,
lookback=lookback,
lr=0.001,
epochs=epochs,
**kwargs)
_ = model.fit(indices='random')
_, pred_y = model.predict(use_datetime_index=False)
return pred_y
def make_and_run(
model,
lookback=12,
epochs=4,
batch_size=16,
data_type='other',
**kwargs):
if data_type == "nasdaq":
df = load_nasdaq()
in_cols = list(df.columns)[0:-1]
inputs = in_cols
outputs = ["NDX"]
else:
outputs = ['blaTEM_coppml']
df = arg_beach(input_features, target=outputs)
inputs = input_features
# column in dataframe to bse used as output/target
df['blaTEM_coppml'] = np.log10(df['blaTEM_coppml'])
kwargs['intervals'] = ((0, 146,),
(145, 386,),
(385, 628,),
(625, 821,),
(821, 1110),
(1110, 1447))
model = model(
data=df,
verbosity=0,
batch_size=batch_size,
lookback=lookback,
lr=0.001,
inputs=inputs,
outputs = outputs,
epochs=epochs,
model={'layers': kwargs.pop('layers')},
**kwargs
)
_ = model.fit(indices='random')
_, pred_y = model.predict(use_datetime_index=False)
return pred_y
import os
import pandas as pd
from inspect import getsourcefile
from os.path import abspath
import site # so that AI4Water directory is in path
site.addsitedir(os.path.dirname(os.path.dirname(__file__)))
from AI4Water.pytorch_models import HARHNModel
from AI4Water.utils.datasets import load_nasdaq
lookback = 10
epochs = 50
df = load_nasdaq()
model = HARHNModel(
batch_size=16,
lookback=lookback,
lr=0.001,
use_predicted_output=True,
intervals=((
0,
146,
), (
145,
386,
), (
385,
628,
), (
625,
821,
