VictorMylle/Thesis
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Rewrote dataset to be able to include new features

Browse Source
This commit is contained in:
Victor Mylle
2023-11-08 23:17:47 +00:00
parent 56c763a6f4
commit 2f48363292
10 changed files with 311218 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
Result-Reports/November.md → Result-Reports/November_1.md
View File

@@ -1,10 +1,11 @@
# Result Report November # Result Report November (1)
## 1. TODOs ## 1. TODOs
- [x] Compare autoregressive vs non-autoregressive - [x] Compare autoregressive vs non-autoregressive
- [ ] Add more input parameters (load forecast) - [ ] Add more input parameters (load forecast)
- [x] Quantile Regression sampling fix - [x] Quantile Regression sampling fix
- [x] Quantile Regression exploration - [x] Quantile Regression exploration
- [ ] Plots with good scaling (y-axis)
## 2. Autoregressive vs Non-Autoregressive ## 2. Autoregressive vs Non-Autoregressive
@@ -85,3 +86,9 @@ Hidden Units: 1024
| --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- |
| 0.1, 0.2, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.8, 0.9 | 65.6542529596313 | 7392.5142575554955 | 77.55779692831604 | 10769.161724849037 | | 0.1, 0.2, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.8, 0.9 | 65.6542529596313 | 7392.5142575554955 | 77.55779692831604 | 10769.161724849037 |
| 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 | 65.68495924348356 | 7326.2239225611975 | 77.62433888969542 | 10789.003223366473 | | 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 | 65.68495924348356 | 7326.2239225611975 | 77.62433888969542 | 10789.003223366473 |
Also tried one with dropout of 0.3, results are better. This needs to be tested more (hyperparameter tuning is not the focus here):
| Quantiles | Train-MAE | Train-MSE | Test-MAE | Test-MSE |
| --- | --- | --- | --- | --- |
| 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 | 63.710736942371994 | 6988.6436956508105 | 77.29499496466444 | 10706.484005597813 |

310945
data/load_forecast.csv Normal file
View File

File diff suppressed because it is too large Load Diff

2
src/data/__init__.py
View File

@@ -1,2 +1,2 @@
from .dataset import NrvDataset from .dataset import NrvDataset
from .preprocessing import DataProcessor from .preprocessing import DataProcessor, DataConfig

56
src/data/dataset.py
View File

@@ -3,13 +3,61 @@ from torch.utils.data import Dataset, DataLoader
import pandas as pd import pandas as pd
class NrvDataset(Dataset): class NrvDataset(Dataset):
def __init__(self, dataframe, sequence_length=96, predict_sequence_length=96): def __init__(self, dataframe, data_config, sequence_length=96, predict_sequence_length=96):
self.data = torch.tensor(dataframe['nrv'].to_numpy(), dtype=torch.float32) self.data_config = data_config
self.nrv = torch.tensor(dataframe['nrv'].to_numpy(), dtype=torch.float32)
self.load_forecast = torch.tensor(dataframe['load_forecast'].to_numpy(), dtype=torch.float32)
self.sequence_length = sequence_length self.sequence_length = sequence_length
self.predict_sequence_length = predict_sequence_length self.predict_sequence_length = predict_sequence_length
def __len__(self): def __len__(self):
return len(self.data) - self.sequence_length - self.predict_sequence_length return len(self.nrv) - self.sequence_length - self.predict_sequence_length
def __getitem__(self, idx): def __getitem__(self, idx):
return self.data[idx:idx+self.sequence_length], self.data[idx+self.sequence_length:idx+self.sequence_length+self.predict_sequence_length] features = []
if self.data_config.NRV_HISTORY:
nrv = self.nrv[idx:idx+self.sequence_length]
features.append(nrv.view(-1))
if self.data_config.LOAD_HISTORY:
load_history = self.load_forecast[idx:idx+self.sequence_length]
features.append(load_history.view(-1))
if self.data_config.LOAD_FORECAST:
load_forecast = self.load_forecast[idx+self.sequence_length:idx+self.sequence_length+self.predict_sequence_length]
features.append(load_forecast.view(-1))
if not features:
raise ValueError("No features are configured to be included in the dataset.")
# Concatenate along dimension 0 to create a one-dimensional feature vector
all_features = torch.cat(features, dim=0)
# Target sequence, flattened if necessary
nrv_target = self.nrv[idx+self.sequence_length:idx+self.sequence_length+self.predict_sequence_length].view(-1)
return all_features, nrv_target
def random_day_autoregressive(self, idx: int):
features = []
# we already have the NRV history with the newly predicted values, so we don't need to include the last 96 values
if self.data_config.LOAD_HISTORY:
load_history = self.load_forecast[idx:idx+self.sequence_length]
features.append(load_history.view(-1))
if self.data_config.LOAD_FORECAST:
load_forecast = self.load_forecast[idx+self.sequence_length:idx+self.sequence_length+self.predict_sequence_length]
features.append(load_forecast.view(-1))
target = self.nrv[idx+self.sequence_length:idx+self.sequence_length+self.predict_sequence_length]
if len(features) == 0:
return None, target
all_features = torch.cat(features, dim=0)
return all_features, target

67
src/data/preprocessing.py
View File

@@ -8,9 +8,15 @@ import pytz
history_data_path = "../../data/history-quarter-hour-data.csv" history_data_path = "../../data/history-quarter-hour-data.csv"
forecast_data_path = "../../data/load_forecast.csv"
class DataConfig:
NRV_HISTORY: bool = True
LOAD_FORECAST: bool = True
LOAD_HISTORY: bool = False
class DataProcessor: class DataProcessor:
def __init__(self): def __init__(self, data_config: DataConfig):
self.batch_size = 2048 self.batch_size = 2048
self.train_range = (-np.inf, datetime(year=2022, month=11, day=30, tzinfo=pytz.UTC)) self.train_range = (-np.inf, datetime(year=2022, month=11, day=30, tzinfo=pytz.UTC))
@@ -18,10 +24,15 @@ class DataProcessor:
self.update_range_str() self.update_range_str()
self.features = ['nrv'] self.history_features = self.get_nrv_history()
self.future_features = self.get_load_forecast()
self.all_features = self.history_features.merge(self.future_features, on='datetime', how='left')
self.data_config = data_config
self.nrv_df = self.get_nrv_history()
self.nrv_scaler = MinMaxScaler(feature_range=(-1, 1)) self.nrv_scaler = MinMaxScaler(feature_range=(-1, 1))
self.load_forecast_scaler = MinMaxScaler(feature_range=(-1, 1))
def set_train_range(self, train_range: tuple): def set_train_range(self, train_range: tuple):
@@ -49,6 +60,22 @@ class DataProcessor:
df.sort_values(by="datetime", inplace=True) df.sort_values(by="datetime", inplace=True)
return df return df
def get_load_forecast(self):
df = pd.read_csv(forecast_data_path, delimiter=';')
df = df.rename(columns={'Day-ahead 6PM forecast': 'load_forecast', 'Datetime': 'datetime'})
df = df[['datetime', 'load_forecast']]
df['datetime'] = pd.to_datetime(df['datetime'], utc=True)
# check if there are nan values
# if df.isnull().values.any():
# # print the rows with nan values
# print(df[df.isnull().any(axis=1)])
# raise ValueError("There are nan values in the load forecast data.")
df.sort_values(by="datetime", inplace=True)
return df
def set_batch_size(self, batch_size: int): def set_batch_size(self, batch_size: int):
self.batch_size = batch_size self.batch_size = batch_size
@@ -56,7 +83,7 @@ class DataProcessor:
return torch.utils.data.DataLoader(dataset, batch_size=self.batch_size, shuffle=shuffle, num_workers=4) return torch.utils.data.DataLoader(dataset, batch_size=self.batch_size, shuffle=shuffle, num_workers=4)
def get_train_dataloader(self, transform: bool = True, predict_sequence_length: int = 96): def get_train_dataloader(self, transform: bool = True, predict_sequence_length: int = 96):
train_df = self.nrv_df.copy() train_df = self.all_features.copy()
if self.train_range[0] != -np.inf: if self.train_range[0] != -np.inf:
train_df = train_df[(train_df['datetime'] >= self.train_range[0])] train_df = train_df[(train_df['datetime'] >= self.train_range[0])]
@@ -66,13 +93,14 @@ class DataProcessor:
if transform: if transform:
train_df['nrv'] = self.nrv_scaler.fit_transform(train_df['nrv'].values.reshape(-1, 1)).reshape(-1) train_df['nrv'] = self.nrv_scaler.fit_transform(train_df['nrv'].values.reshape(-1, 1)).reshape(-1)
train_df['load_forecast'] = self.load_forecast_scaler.fit_transform(train_df['load_forecast'].values.reshape(-1, 1)).reshape(-1)
train_dataset = NrvDataset(train_df, predict_sequence_length=predict_sequence_length) train_dataset = NrvDataset(train_df, data_config=self.data_config, predict_sequence_length=predict_sequence_length)
return self.get_dataloader(train_dataset) return self.get_dataloader(train_dataset)
def get_test_dataloader(self, transform: bool = True, predict_sequence_length: int = 96): def get_test_dataloader(self, transform: bool = True, predict_sequence_length: int = 96):
test_df = self.nrv_df.copy() test_df = self.all_features.copy()
if self.test_range[0] != -np.inf: if self.test_range[0] != -np.inf:
test_df = test_df[(test_df['datetime'] >= self.test_range[0])] test_df = test_df[(test_df['datetime'] >= self.test_range[0])]
@@ -82,35 +110,14 @@ class DataProcessor:
if transform: if transform:
test_df['nrv'] = self.nrv_scaler.transform(test_df['nrv'].values.reshape(-1, 1)).reshape(-1) test_df['nrv'] = self.nrv_scaler.transform(test_df['nrv'].values.reshape(-1, 1)).reshape(-1)
test_dataset = NrvDataset(test_df, predict_sequence_length=predict_sequence_length) test_df['load_forecast'] = self.load_forecast_scaler.transform(test_df['load_forecast'].values.reshape(-1, 1)).reshape(-1)
test_dataset = NrvDataset(test_df, data_config=self.data_config, predict_sequence_length=predict_sequence_length)
return self.get_dataloader(test_dataset, shuffle=False) return self.get_dataloader(test_dataset, shuffle=False)
def get_dataloaders(self, transform: bool = True, predict_sequence_length: int = 96): def get_dataloaders(self, transform: bool = True, predict_sequence_length: int = 96):
return self.get_train_dataloader(transform=transform, predict_sequence_length=predict_sequence_length), self.get_test_dataloader(transform=transform, predict_sequence_length=predict_sequence_length) return self.get_train_dataloader(transform=transform, predict_sequence_length=predict_sequence_length), self.get_test_dataloader(transform=transform, predict_sequence_length=predict_sequence_length)
def get_random_day(self, train: bool = True, transform: bool = True):
df = self.nrv_df.copy()
range = self.train_range if train else self.test_range
if range[0] != -np.inf:
df = df[(df['datetime'] >= range[0])]
if range[1] != np.inf:
df = df[(df['datetime'] <= range[1])]
if transform:
df['nrv'] = self.nrv_scaler.transform(df['nrv'].values.reshape(-1, 1)).reshape(-1)
data_tensor = torch.tensor(df[self.features].values, dtype=torch.float32)
random_start_idx = np.random.randint(0, len(df) - 191)
current_day_features = data_tensor[random_start_idx:random_start_idx+96]
next_day_features = data_tensor[random_start_idx+96:random_start_idx+192]
return (current_day_features, next_day_features)
def inverse_transform(self, tensor: torch.Tensor): def inverse_transform(self, tensor: torch.Tensor):
return self.nrv_scaler.inverse_transform(tensor.cpu().numpy()).reshape(-1) return self.nrv_scaler.inverse_transform(tensor.cpu().numpy()).reshape(-1)

6
src/models/non_linear_regression.py
View File

@@ -1,7 +1,7 @@
import torch import torch
class NonLinearRegression(torch.nn.Module): class NonLinearRegression(torch.nn.Module):
def __init__(self, inputSize, output_size, hiddenSize=128, numLayers=2): def __init__(self, inputSize, output_size, hiddenSize=128, numLayers=2, dropout=0.0):
super(NonLinearRegression, self).__init__() super(NonLinearRegression, self).__init__()
self.inputSize = inputSize self.inputSize = inputSize
self.output_size = output_size self.output_size = output_size
@@ -9,11 +9,15 @@ class NonLinearRegression(torch.nn.Module):
self.hiddenSize = hiddenSize self.hiddenSize = hiddenSize
self.numLayers = numLayers self.numLayers = numLayers
self.dropout = dropout
# add linear layers with relu # add linear layers with relu
self.layers = torch.nn.ModuleList() self.layers = torch.nn.ModuleList()
self.layers.append(torch.nn.Linear(inputSize, hiddenSize)) self.layers.append(torch.nn.Linear(inputSize, hiddenSize))
self.layers.append(torch.nn.Dropout(dropout))
for _ in range(numLayers - 2): for _ in range(numLayers - 2):
self.layers.append(torch.nn.Linear(hiddenSize, hiddenSize)) self.layers.append(torch.nn.Linear(hiddenSize, hiddenSize))
self.layers.append(torch.nn.Dropout(dropout))
self.layers.append(torch.nn.Linear(hiddenSize, output_size)) self.layers.append(torch.nn.Linear(hiddenSize, output_size))
self.relu = torch.nn.ReLU() self.relu = torch.nn.ReLU()

99
src/notebooks/training.ipynb
View File

@@ -8,7 +8,7 @@
"source": [ "source": [
"import sys\n", "import sys\n",
"sys.path.append('..')\n", "sys.path.append('..')\n",
"from data import DataProcessor\n", "from data import DataProcessor, DataConfig\n",
"from trainers.quantile_trainer import QuantileTrainer\n", "from trainers.quantile_trainer import QuantileTrainer\n",
"from trainers.autoregressive_trainer import AutoRegressiveTrainer\n", "from trainers.autoregressive_trainer import AutoRegressiveTrainer\n",
"from trainers.trainer import Trainer\n", "from trainers.trainer import Trainer\n",
@@ -50,35 +50,40 @@
"name": "stdout", "name": "stdout",
"output_type": "stream", "output_type": "stream",
"text": [ "text": [
"ClearML Task: created new task id=909da25a8d214f75ab3858506ae615e8\n", "ClearML Task: created new task id=ae9d93e6bf3f4d228832378707574777\n",
"2023-11-07 16:29:35,665 - clearml.Task - INFO - Storing jupyter notebook directly as code\n", "ClearML results page: http://192.168.1.182:8080/projects/2e46d4af6f1e4c399cf9f5aa30bc8795/experiments/ae9d93e6bf3f4d228832378707574777/output/log\n",
"ClearML results page: http://192.168.1.182:8080/projects/2e46d4af6f1e4c399cf9f5aa30bc8795/experiments/909da25a8d214f75ab3858506ae615e8/output/log\n", "2023-11-08 21:46:32,338 - clearml.Task - INFO - Storing jupyter notebook directly as code\n",
"2023-11-07 16:30:08,121 - clearml.model - WARNING - 500 model found when searching for `file:///workspaces/Thesis/src/notebooks/checkpoint.pt`\n", "2023-11-08 21:46:32,342 - clearml.Repository Detection - WARNING - Can't get url information for git repo in /workspaces/Thesis/src/notebooks\n",
"2023-11-07 16:30:08,123 - clearml.model - WARNING - Selected model `Quantile Regression - Linear` (id=bc0cb0d7fc614e2e8b0edf5b85348646)\n", "2023-11-08 21:46:35,244 - clearml.model - WARNING - 500 model found when searching for `file:///workspaces/Thesis/src/notebooks/checkpoint.pt`\n",
"2023-11-07 16:30:08,130 - clearml.frameworks - INFO - Found existing registered model id=bc0cb0d7fc614e2e8b0edf5b85348646 [/workspaces/Thesis/src/notebooks/checkpoint.pt] reusing it.\n", "2023-11-08 21:46:35,245 - clearml.model - WARNING - Selected model `Untitled Task` (id=bc0cb0d7fc614e2e8b0edf5b85348646)\n",
"2023-11-07 16:30:08,677 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Non-AutoRegressive%20-%20Non%20Linear%20%283%20hidden%20layers%20-%201024%20units%29.909da25a8d214f75ab3858506ae615e8/models/checkpoint.pt\n", "2023-11-08 21:46:35,254 - clearml.frameworks - INFO - Found existing registered model id=bc0cb0d7fc614e2e8b0edf5b85348646 [/workspaces/Thesis/src/notebooks/checkpoint.pt] reusing it.\n",
"2023-11-07 16:30:10,302 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Non-AutoRegressive%20-%20Non%20Linear%20%283%20hidden%20layers%20-%201024%20units%29.909da25a8d214f75ab3858506ae615e8/models/checkpoint.pt\n", "2023-11-08 21:46:35,748 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.ae9d93e6bf3f4d228832378707574777/models/checkpoint.pt\n",
"2023-11-08 21:46:37,673 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.ae9d93e6bf3f4d228832378707574777/models/checkpoint.pt\n",
"2023-11-08 21:46:41,215 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.ae9d93e6bf3f4d228832378707574777/models/checkpoint.pt\n",
"2023-11-08 21:46:43,004 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.ae9d93e6bf3f4d228832378707574777/models/checkpoint.pt\n",
"2023-11-08 21:46:44,745 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.ae9d93e6bf3f4d228832378707574777/models/checkpoint.pt\n",
"Early stopping triggered\n" "Early stopping triggered\n"
] ]
} }
], ],
"source": [ "source": [
"#### Hyperparameters ####\n", "#### Hyperparameters ####\n",
"inputDim = 96\n", "inputDim = 96 + 96\n",
"learningRate = 0.0003\n", "learningRate = 0.0003\n",
"epochs = 50\n", "epochs = 150\n",
"\n", "\n",
"# model = LinearRegression(inputDim, 96)\n", "# model = LinearRegression(inputDim, 96)\n",
"model = NonLinearRegression(inputDim, 96, hiddenSize=1024, numLayers=5)\n", "model = NonLinearRegression(inputDim, 96, hiddenSize=1024, numLayers=5)\n",
"optimizer = torch.optim.Adam(model.parameters(), lr=learningRate)\n", "optimizer = torch.optim.Adam(model.parameters(), lr=learningRate)\n",
"\n", "\n",
"#### Data Processor ####\n", "#### Data Processor ####\n",
"data_processor = DataProcessor()\n", "data_config = DataConfig()\n",
"data_processor = DataProcessor(data_config)\n",
"data_processor.set_batch_size(1024)\n", "data_processor.set_batch_size(1024)\n",
"\n", "\n",
"\n", "\n",
"data_processor.set_train_range((datetime(year=2015, month=1, day=1, tzinfo=pytz.UTC), datetime(year=2022, month=11, day=30, tzinfo=pytz.UTC)))\n", "data_processor.set_train_range((datetime(year=2015, month=1, day=1, tzinfo=pytz.UTC), datetime(year=2022, month=11, day=30, tzinfo=pytz.UTC)))\n",
"data_processor.set_test_range((datetime(year=2023, month=1, day=1, tzinfo=pytz.UTC), np.inf))\n", "data_processor.set_test_range((datetime(year=2023, month=1, day=1, tzinfo=pytz.UTC), datetime(year=2023, month=6, day=20, tzinfo=pytz.UTC)))\n",
"\n", "\n",
"#### ClearML ####\n", "#### ClearML ####\n",
"clearml_helper = ClearMLHelper(project_name=\"Thesis/NrvForecast\")\n", "clearml_helper = ClearMLHelper(project_name=\"Thesis/NrvForecast\")\n",
@@ -87,7 +92,7 @@
"trainer = Trainer(model, optimizer, nn.MSELoss(), data_processor, \"cuda\", debug=False, clearml_helper=clearml_helper)\n", "trainer = Trainer(model, optimizer, nn.MSELoss(), data_processor, \"cuda\", debug=False, clearml_helper=clearml_helper)\n",
"trainer.add_metrics_to_track([MSELoss(), L1Loss()])\n", "trainer.add_metrics_to_track([MSELoss(), L1Loss()])\n",
"trainer.plot_every(10)\n", "trainer.plot_every(10)\n",
"trainer.early_stopping(patience=10)\n", "trainer.early_stopping(patience=40)\n",
"trainer.train(epochs=epochs)" "trainer.train(epochs=epochs)"
] ]
}, },
@@ -100,15 +105,30 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 3, "execution_count": 2,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"InsecureRequestWarning: Certificate verification is disabled! Adding certificate verification is strongly advised. See: https://urllib3.readthedocs.io/en/latest/advanced-usage.html#ssl-warnings\n"
]
},
{ {
"name": "stdout", "name": "stdout",
"output_type": "stream", "output_type": "stream",
"text": [ "text": [
"ClearML Task: created new task id=6467cef37fdc408d95b89f0dca0e26dd\n", "ClearML Task: created new task id=e4983af7aad748e8ae71e5fa46767d60\n",
"ClearML results page: http://192.168.1.182:8080/projects/2e46d4af6f1e4c399cf9f5aa30bc8795/experiments/6467cef37fdc408d95b89f0dca0e26dd/output/log\n", "ClearML results page: http://192.168.1.182:8080/projects/2e46d4af6f1e4c399cf9f5aa30bc8795/experiments/e4983af7aad748e8ae71e5fa46767d60/output/log\n",
"2023-11-08 22:51:38,769 - clearml.Task - INFO - Storing jupyter notebook directly as code\n",
"2023-11-08 22:51:38,773 - clearml.Repository Detection - WARNING - Can't get url information for git repo in /workspaces/Thesis/src/notebooks\n",
"2023-11-08 22:51:41,704 - clearml.model - WARNING - 500 model found when searching for `file:///workspaces/Thesis/src/notebooks/checkpoint.pt`\n",
"2023-11-08 22:51:41,705 - clearml.model - WARNING - Selected model `Untitled Task` (id=bc0cb0d7fc614e2e8b0edf5b85348646)\n",
"2023-11-08 22:51:41,713 - clearml.frameworks - INFO - Found existing registered model id=bc0cb0d7fc614e2e8b0edf5b85348646 [/workspaces/Thesis/src/notebooks/checkpoint.pt] reusing it.\n",
"2023-11-08 22:51:42,239 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.e4983af7aad748e8ae71e5fa46767d60/models/checkpoint.pt\n",
"2023-11-08 22:51:43,980 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.e4983af7aad748e8ae71e5fa46767d60/models/checkpoint.pt\n",
"2023-11-08 22:51:45,684 - clearml.Task - INFO - Completed model upload to http://192.168.1.182:8081/Thesis/NrvForecast/Untitled%20Task.e4983af7aad748e8ae71e5fa46767d60/models/checkpoint.pt\n",
"Early stopping triggered\n" "Early stopping triggered\n"
] ]
} }
@@ -124,7 +144,9 @@
"optimizer = torch.optim.Adam(model.parameters(), lr=learningRate)\n", "optimizer = torch.optim.Adam(model.parameters(), lr=learningRate)\n",
"\n", "\n",
"#### Data Processor ####\n", "#### Data Processor ####\n",
"data_processor = DataProcessor()\n", "data_config = DataConfig()\n",
"data_config.LOAD_FORECAST = False\n",
"data_processor = DataProcessor(data_config)\n",
"data_processor.set_batch_size(1024)\n", "data_processor.set_batch_size(1024)\n",
"\n", "\n",
"\n", "\n",
@@ -151,26 +173,18 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 4, "execution_count": 2,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
"name": "stderr", "ename": "TypeError",
"output_type": "stream", "evalue": "DataProcessor.__init__() missing 1 required positional argument: 'data_config'",
"text": [ "output_type": "error",
"/workspaces/Thesis/src/notebooks/../trainers/quantile_trainer.py:16: UserWarning:\n", "traceback": [
"\n", "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n", "\u001b[0;31mTypeError\u001b[0m Traceback (most recent call last)",
"\n" "\u001b[1;32m/workspaces/Thesis/src/notebooks/training.ipynb Cell 7\u001b[0m line \u001b[0;36m1\n\u001b[1;32m <a href='vscode-notebook-cell://dev-container%2B7b22686f737450617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f546865736973222c226c6f63616c446f636b6572223a66616c73652c22636f6e66696746696c65223a7b22246d6964223a312c2270617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f5468657369732f2e646576636f6e7461696e65722f646576636f6e7461696e65722e6a736f6e222c22736368656d65223a227673636f64652d66696c65486f7374227d7d@ssh-remote%2Bvictormylle.be/workspaces/Thesis/src/notebooks/training.ipynb#X10sdnNjb2RlLXJlbW90ZQ%3D%3D?line=13'>14</a>\u001b[0m data_config \u001b[39m=\u001b[39m DataConfig()\n\u001b[1;32m <a href='vscode-notebook-cell://dev-container%2B7b22686f737450617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f546865736973222c226c6f63616c446f636b6572223a66616c73652c22636f6e66696746696c65223a7b22246d6964223a312c2270617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f5468657369732f2e646576636f6e7461696e65722f646576636f6e7461696e65722e6a736f6e222c22736368656d65223a227673636f64652d66696c65486f7374227d7d@ssh-remote%2Bvictormylle.be/workspaces/Thesis/src/notebooks/training.ipynb#X10sdnNjb2RlLXJlbW90ZQ%3D%3D?line=14'>15</a>\u001b[0m data_config\u001b[39m.\u001b[39mLOAD_FORECAST \u001b[39m=\u001b[39m \u001b[39mFalse\u001b[39;00m\n\u001b[0;32m---> <a href='vscode-notebook-cell://dev-container%2B7b22686f737450617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f546865736973222c226c6f63616c446f636b6572223a66616c73652c22636f6e66696746696c65223a7b22246d6964223a312c2270617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f5468657369732f2e646576636f6e7461696e65722f646576636f6e7461696e65722e6a736f6e222c22736368656d65223a227673636f64652d66696c65486f7374227d7d@ssh-remote%2Bvictormylle.be/workspaces/Thesis/src/notebooks/training.ipynb#X10sdnNjb2RlLXJlbW90ZQ%3D%3D?line=15'>16</a>\u001b[0m data_processor \u001b[39m=\u001b[39m DataProcessor()\n\u001b[1;32m <a href='vscode-notebook-cell://dev-container%2B7b22686f737450617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f546865736973222c226c6f63616c446f636b6572223a66616c73652c22636f6e66696746696c65223a7b22246d6964223a312c2270617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f5468657369732f2e646576636f6e7461696e65722f646576636f6e7461696e65722e6a736f6e222c22736368656d65223a227673636f64652d66696c65486f7374227d7d@ssh-remote%2Bvictormylle.be/workspaces/Thesis/src/notebooks/training.ipynb#X10sdnNjb2RlLXJlbW90ZQ%3D%3D?line=16'>17</a>\u001b[0m data_processor\u001b[39m.\u001b[39mset_batch_size(\u001b[39m1024\u001b[39m)\n\u001b[1;32m <a href='vscode-notebook-cell://dev-container%2B7b22686f737450617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f546865736973222c226c6f63616c446f636b6572223a66616c73652c22636f6e66696746696c65223a7b22246d6964223a312c2270617468223a222f686f6d652f766963746f726d796c6c652f53656144726976652f4d79204c69627261726965732f4750552d7365727665722f5468657369732f2e646576636f6e7461696e65722f646576636f6e7461696e65722e6a736f6e222c22736368656d65223a227673636f64652d66696c65486f7374227d7d@ssh-remote%2Bvictormylle.be/workspaces/Thesis/src/notebooks/training.ipynb#X10sdnNjb2RlLXJlbW90ZQ%3D%3D?line=18'>19</a>\u001b[0m data_processor\u001b[39m.\u001b[39mset_train_range((\u001b[39m-\u001b[39mnp\u001b[39m.\u001b[39minf, datetime(year\u001b[39m=\u001b[39m\u001b[39m2022\u001b[39m, month\u001b[39m=\u001b[39m\u001b[39m11\u001b[39m, day\u001b[39m=\u001b[39m\u001b[39m30\u001b[39m, tzinfo\u001b[39m=\u001b[39mpytz\u001b[39m.\u001b[39mUTC)))\n",
] "\u001b[0;31mTypeError\u001b[0m: DataProcessor.__init__() missing 1 required positional argument: 'data_config'"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"ClearML Task: created new task id=07a2dc72793446d8a8101eafce0d80db\n",
"ClearML results page: http://192.168.1.182:8080/projects/2e46d4af6f1e4c399cf9f5aa30bc8795/experiments/07a2dc72793446d8a8101eafce0d80db/output/log\n",
"Early stopping triggered\n"
] ]
} }
], ],
@@ -178,17 +192,19 @@
"#### Hyperparameters ####\n", "#### Hyperparameters ####\n",
"inputDim = 96\n", "inputDim = 96\n",
"learningRate = 0.0003\n", "learningRate = 0.0003\n",
"epochs = 50\n", "epochs = 100\n",
"\n", "\n",
"quantiles = torch.tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]).to(\"cuda\")\n", "quantiles = torch.tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]).to(\"cuda\")\n",
"\n", "\n",
"# model = LinearRegression(inputDim, len(quantiles))\n", "# model = LinearRegression(inputDim, len(quantiles))\n",
"model = NonLinearRegression(inputDim, len(quantiles), hiddenSize=1024, numLayers=5)\n", "model = NonLinearRegression(inputDim, len(quantiles), hiddenSize=1024, numLayers=5, dropout=0.2)\n",
"model.output_size = 1\n", "model.output_size = 1\n",
"optimizer = torch.optim.Adam(model.parameters(), lr=learningRate)\n", "optimizer = torch.optim.Adam(model.parameters(), lr=learningRate)\n",
"\n", "\n",
"#### Data Processor ####\n", "#### Data Processor ####\n",
"data_processor = DataProcessor()\n", "data_config = DataConfig()\n",
"data_config.LOAD_FORECAST = False\n",
"data_processor = DataProcessor(data_config)\n",
"data_processor.set_batch_size(1024)\n", "data_processor.set_batch_size(1024)\n",
"\n", "\n",
"data_processor.set_train_range((-np.inf, datetime(year=2022, month=11, day=30, tzinfo=pytz.UTC)))\n", "data_processor.set_train_range((-np.inf, datetime(year=2022, month=11, day=30, tzinfo=pytz.UTC)))\n",
@@ -204,6 +220,13 @@
"trainer.plot_every(10)\n", "trainer.plot_every(10)\n",
"trainer.train(epochs=epochs)" "trainer.train(epochs=epochs)"
] ]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
} }
], ],
"metadata": { "metadata": {

53
src/trainers/autoregressive_trainer.py
View File

@@ -10,19 +10,16 @@ from plotly.subplots import make_subplots
from trainers.trainer import Trainer from trainers.trainer import Trainer
class AutoRegressiveTrainer(Trainer): class AutoRegressiveTrainer(Trainer):
def debug_plots(self, task, train: bool, samples, epoch): def debug_plots(self, task, train: bool, data_loader, sample_indices, epoch):
X, y = samples num_samples = len(sample_indices)
X = X.to(self.device)
num_samples = len(X)
rows = num_samples # One row per sample since we only want one column rows = num_samples # One row per sample since we only want one column
cols = 1 cols = 1
fig = make_subplots(rows=rows, cols=cols, subplot_titles=[f'Sample {i+1}' for i in range(num_samples)]) fig = make_subplots(rows=rows, cols=cols, subplot_titles=[f'Sample {i+1}' for i in range(num_samples)])
for i, (current_day, next_day) in enumerate(zip(X, y)): for i, idx in enumerate(sample_indices):
predictions = self.predict_auto_regressive(current_day) initial, predictions, target = self.auto_regressive(data_loader, idx)
sub_fig = self.get_plot(current_day, next_day, predictions, show_legend=(i == 0)) sub_fig = self.get_plot(initial, target, predictions, show_legend=(i == 0))
row = i + 1 row = i + 1
col = 1 col = 1
@@ -30,7 +27,7 @@ class AutoRegressiveTrainer(Trainer):
for trace in sub_fig.data: for trace in sub_fig.data:
fig.add_trace(trace, row=row, col=col) fig.add_trace(trace, row=row, col=col)
loss = self.criterion(predictions.to(self.device), next_day.to(self.device)).item() loss = self.criterion(predictions.to(self.device), target.to(self.device)).item()
fig['layout']['annotations'][i].update(text=f"{loss.__class__.__name__}: {loss:.6f}") fig['layout']['annotations'][i].update(text=f"{loss.__class__.__name__}: {loss:.6f}")
@@ -46,14 +43,38 @@ class AutoRegressiveTrainer(Trainer):
figure=fig figure=fig
) )
def auto_regressive(self, data_loader, idx, sequence_length: int = 96):
self.model.eval()
target_full = []
predictions_full = []
def predict_auto_regressive(self, initial_sequence: torch.Tensor, sequence_length: int = 96): prev_features, target = data_loader.dataset[idx]
initial_sequence = initial_sequence.to(self.device) prev_features = prev_features.to(self.device)
return predict_auto_regressive(self.model, initial_sequence, sequence_length) initial_sequence = prev_features[:96]
def random_day_prediction(self): target_full.append(target)
current_day_features, next_day_features = self.data_processor.get_random_test_day() with torch.no_grad():
prediction = self.model(prev_features.unsqueeze(0))
predictions_full.append(prediction.squeeze(-1))
predictions = self.predict_auto_regressive(current_day_features) for i in range(sequence_length - 1):
return current_day_features, next_day_features, predictions new_features = torch.cat((prev_features[1:97].cpu(), prediction.squeeze(-1).cpu()), dim=0)
# get the other needed features
other_features, new_target = data_loader.dataset.random_day_autoregressive(idx + i + 1)
if other_features is not None:
prev_features = torch.cat((new_features, other_features), dim=0)
else:
prev_features = new_features
# add target to target_full
target_full.append(new_target)
# predict
with torch.no_grad():
prediction = self.model(new_features.unsqueeze(0).to(self.device))
predictions_full.append(prediction.squeeze(-1))
return initial_sequence.cpu(), torch.stack(predictions_full).cpu(), torch.stack(target_full).cpu()

50
src/trainers/quantile_trainer.py
View File

@@ -20,10 +20,6 @@ class QuantileTrainer(AutoRegressiveTrainer):
criterion = PinballLoss(quantiles=quantiles_tensor) criterion = PinballLoss(quantiles=quantiles_tensor)
super().__init__(model=model, optimizer=optimizer, criterion=criterion, data_processor=data_processor, device=device, clearml_helper=clearml_helper, debug=debug) super().__init__(model=model, optimizer=optimizer, criterion=criterion, data_processor=data_processor, device=device, clearml_helper=clearml_helper, debug=debug)
def predict_auto_regressive(self, initial_sequence: torch.Tensor, sequence_length: int = 96):
initial_sequence = initial_sequence.to(self.device)
return predict_auto_regressive_quantile(self.model, self.sample_from_dist, initial_sequence, self.quantiles, sequence_length)
def log_final_metrics(self, task, dataloader, train: bool = True): def log_final_metrics(self, task, dataloader, train: bool = True):
metrics = { metric.__class__.__name__: 0.0 for metric in self.metrics_to_track } metrics = { metric.__class__.__name__: 0.0 for metric in self.metrics_to_track }
@@ -85,6 +81,52 @@ class QuantileTrainer(AutoRegressiveTrainer):
return fig return fig
def auto_regressive(self, data_loader, idx, sequence_length: int = 96):
self.model.eval()
target_full = []
predictions_sampled = []
predictions_full = []
prev_features, target = data_loader.dataset[idx]
prev_features = prev_features.to(self.device)
initial_sequence = prev_features[:96]
target_full.append(target)
with torch.no_grad():
prediction = self.model(prev_features.unsqueeze(0))
predictions_full.append(prediction.squeeze(0))
# sample from the distribution
sample = self.sample_from_dist(self.quantiles.cpu(), prediction.squeeze(-1).cpu().numpy())
predictions_sampled.append(sample)
for i in range(sequence_length - 1):
new_features = torch.cat((prev_features[1:97].cpu(), torch.tensor([predictions_sampled[-1]])), dim=0)
new_features = new_features.float()
# get the other needed features
other_features, new_target = data_loader.dataset.random_day_autoregressive(idx + i + 1)
if other_features is not None:
prev_features = torch.cat((new_features, other_features), dim=0)
else:
prev_features = new_features
# add target to target_full
target_full.append(new_target)
# predict
with torch.no_grad():
prediction = self.model(new_features.unsqueeze(0).to(self.device))
predictions_full.append(prediction.squeeze(0))
# sample from the distribution
sample = self.sample_from_dist(self.quantiles.cpu(), prediction.squeeze(-1).cpu().numpy())
predictions_sampled.append(sample)
return initial_sequence.cpu(), torch.stack(predictions_full).cpu(), torch.stack(target_full).cpu()
@staticmethod @staticmethod
def sample_from_dist(quantiles, output_values): def sample_from_dist(quantiles, output_values):
# Interpolate the inverse CDF # Interpolate the inverse CDF

47
src/trainers/trainer.py
View File

@@ -70,23 +70,22 @@ class Trainer:
return task return task
def random_samples(self, train: bool = True, num_samples: int = 10): def random_samples(self, train: bool = True, num_samples: int = 10):
random_X = [] train_loader, test_loader = self.data_processor.get_dataloaders(predict_sequence_length=self.model.output_size)
random_Y = []
for _ in range(num_samples): if train:
X, y = self.data_processor.get_random_day(train=train) loader = train_loader
random_X.append(X) else:
random_Y.append(y) loader = test_loader
indices = np.random.randint(0, len(loader.dataset) - 1, size=num_samples)
return indices
random_X = torch.stack(random_X)
random_Y = torch.stack(random_Y)
return random_X, random_Y
def train(self, epochs: int): def train(self, epochs: int):
train_loader, test_loader = self.data_processor.get_dataloaders(predict_sequence_length=self.model.output_size) train_loader, test_loader = self.data_processor.get_dataloaders(predict_sequence_length=self.model.output_size)
train_random_X, train_random_y = self.random_samples(train=True) train_samples = self.random_samples(train=True)
test_random_X, test_random_y = self.random_samples(train=False) test_samples = self.random_samples(train=False)
task = self.init_clearml_task() task = self.init_clearml_task()
@@ -129,8 +128,8 @@ class Trainer:
if epoch % self.plot_every_n_epochs == 0: if epoch % self.plot_every_n_epochs == 0:
self.debug_plots(task, True, (train_random_X, train_random_y), epoch) self.debug_plots(task, True, train_loader, train_samples, epoch)
self.debug_plots(task, False, (test_random_X, test_random_y), epoch) self.debug_plots(task, False, test_loader, test_samples, epoch)
if task: if task:
self.finish_training(task=task) self.finish_training(task=task)
@@ -144,6 +143,7 @@ class Trainer:
with torch.no_grad(): with torch.no_grad():
for inputs, targets in dataloader: for inputs, targets in dataloader:
inputs, targets = inputs.to(self.device), targets inputs, targets = inputs.to(self.device), targets
outputs = self.model(inputs) outputs = self.model(inputs)
inversed_outputs = torch.tensor(self.data_processor.inverse_transform(outputs)) inversed_outputs = torch.tensor(self.data_processor.inverse_transform(outputs))
@@ -215,22 +215,25 @@ class Trainer:
return fig return fig
def debug_plots(self, task, train: bool, samples, epoch): def debug_plots(self, task, train: bool, data_loader, sample_indices, epoch):
X, y = samples num_samples = len(sample_indices)
X = X.to(self.device)
num_samples = len(X)
rows = num_samples # One row per sample since we only want one column rows = num_samples # One row per sample since we only want one column
cols = 1 cols = 1
fig = make_subplots(rows=rows, cols=cols, subplot_titles=[f'Sample {i+1}' for i in range(num_samples)]) fig = make_subplots(rows=rows, cols=cols, subplot_titles=[f'Sample {i+1}' for i in range(num_samples)])
for i, (current_day, next_day) in enumerate(zip(X, y)): for i, idx in enumerate(sample_indices):
features, target = data_loader.dataset[idx]
features = features.to(self.device)
target = target.to(self.device)
self.model.eval() self.model.eval()
with torch.no_grad(): with torch.no_grad():
predictions = self.model(current_day).cpu() predictions = self.model(features).cpu()
sub_fig = self.get_plot(current_day, next_day, predictions, show_legend=(i == 0)) sub_fig = self.get_plot(features[:96], target, predictions, show_legend=(i == 0))
row = i + 1 row = i + 1
col = 1 col = 1
@@ -239,7 +242,7 @@ class Trainer:
fig.add_trace(trace, row=row, col=col) fig.add_trace(trace, row=row, col=col)
loss = self.criterion(predictions.to(self.device), next_day.squeeze(-1).to(self.device)).item() loss = self.criterion(predictions.to(self.device), target.squeeze(-1).to(self.device)).item()
fig['layout']['annotations'][i].update(text=f"{loss.__class__.__name__}: {loss:.6f}") fig['layout']['annotations'][i].update(text=f"{loss.__class__.__name__}: {loss:.6f}")