Skip to content

Instantly share code, notes, and snippets.

View bkaankuguoglu's full-sized avatar

Kaan Kuguoglu bkaankuguoglu

15 followers · 12 following
View GitHub Profile
@bkaankuguoglu
bkaankuguoglu / forecast_with_lag_features
Created January 16, 2022 17:45
Class Optimization
# ...
def forecast_with_lag_features(self, test_loader, batch_size=1, n_features=1, n_steps=100):
test_loader_iter = iter(test_loader)
predictions = []
*_, (X, y) = test_loader_iter
y = y.to(device).detach().numpy()
@bkaankuguoglu
bkaankuguoglu / plot_dataset_with_forecast
Created January 15, 2022 15:35
def plot_dataset_with_forecast(df, df_forecast, title):
data = []
value = go.Scatter(
x=df.index,
y=df.value,
mode="lines",
name="values",
marker=dict(),
text=df.index,
@bkaankuguoglu
bkaankuguoglu / format_forecasts
Created January 15, 2022 15:05
def format_forecasts(forecasts, index, scaler):
preds = np.concatenate(forecasts, axis=0).ravel()
df_forecast = pd.DataFrame(data={"prediction": preds}, index=index)
df_result = df_forecast.sort_index()
df_result = inverse_transform(scaler, df_result, [["prediction"]])
return df_result
df_forecast = format_forecasts(forecasts, index, scaler)
@bkaankuguoglu
bkaankuguoglu / opt_forecast_with_predictors
Created January 15, 2022 15:04
forecasts = opt.forecast_with_predictors(forecast_loader,
batch_size=1,
n_features=input_dim,
n_steps=100)
@bkaankuguoglu
bkaankuguoglu / create_forecast_loader
Created January 15, 2022 15:01
X_forecast, y_forecast = feature_label_split(df_forecast, 'value')
scaler = get_scaler('minmax')
X_train_arr = scaler.fit_transform(X_train)
X_forecast_arr = scaler.transform(X_forecast)
y_train_arr = scaler.fit_transform(y_train)
y_forecast_arr = scaler.transform(y_forecast)
forecast_dataset = TensorDataset(torch.Tensor(X_forecast_arr),
torch.Tensor(y_forecast_arr))
@bkaankuguoglu
bkaankuguoglu / generate_future_features
Created November 25, 2021 15:01
df_forecast['value'] = 0
df_forecast= (df_forecast
.assign(hour = df_forecast.index.hour)
.assign(day = df_forecast.index.day)
.assign(month = df_forecast.index.month)
.assign(day_of_week = df_forecast.index.dayofweek)
.assign(week_of_year = df_forecast.index.week)
)
df_forecast = onehot_encode_pd(df_forecast, ['month','day','day_of_week','week_of_year'])
@bkaankuguoglu
bkaankuguoglu / generate_forecast_index
Created November 25, 2021 14:58
def get_datetime_index(df):
return (
pd.to_datetime(df.index[-1])
+ (pd.to_datetime(df.index[-1]) - pd.to_datetime(df.index[-2])),
pd.to_datetime(df.index[-1]) - pd.to_datetime(df.index[-2]),
)
start_date, freq = get_datetime_index(y_test)
index = pd.date_range(start=start_date, freq=freq, periods=100)
df_forecast = pd.DataFrame(index=index)
@bkaankuguoglu
bkaankuguoglu / forecast_with_predictors
Last active January 15, 2022 12:30
class Optimization:
# ...
def forecast_with_predictors(
self, forecast_loader, batch_size=1, n_features=1, n_steps=100
):
"""Forecasts values for RNNs with predictors and one-dimensional output
The method takes DataLoader for the test dataset, batch size for mini-batch testing,
number of features and number of steps to predict as inputs. Then it generates the
@bkaankuguoglu
bkaankuguoglu / build_baseline_model
Created May 15, 2021 14:13
from sklearn.linear_model import LinearRegression
def build_baseline_model(df, test_ratio, target_col):
X, y = feature_label_split(df, target_col)
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=test_ratio, shuffle=False
)
model = LinearRegression()
model.fit(X_train, y_train)
prediction = model.predict(X_test)
@bkaankuguoglu
bkaankuguoglu / plot_predictions
Last active May 15, 2021 14:16
import plotly.offline as pyo
def plot_predictions(df_result, df_baseline):
data = []
value = go.Scatter(
x=df_result.index,
y=df_result.value,
mode="lines",
name="values",