tortreinador.utils package

Submodules

tortreinador.utils.Recorder module

class tortreinador.utils.Recorder.Recorder(device)

Bases: object

avg()

reset()

update(val)

class tortreinador.utils.Recorder.RecorderForEpoch(device)

Bases: object

avg()

reset()

update(val)

tortreinador.utils.View module

tortreinador.utils.View.get_lastlayer_params(net)

get last trainable layer of a net :param network architectur:

Returns:

last layer weights and last layer bias

tortreinador.utils.View.init_weights(net)

the weights of conv layer and fully connected layers are both initilized with Xavier algorithm, In particular, we set the parameters to random values uniformly drawn from [-a, a] where a = sqrt(6 * (din + dout)), for batch normalization layers, y=1, b=0, all bias initialized to 0.

tortreinador.utils.View.split_weights(net)

split network weights into to categlories, one are weights in conv layer and linear layer, others are other learnable paramters(conv bias, bn weights, bn bias, linear bias) :param net: network architecture

Returns:

a dictionary of params splite into to categlories

tortreinador.utils.View.visualize_lastlayer(writer, net, n_iter)

visualize last layer grads

tortreinador.utils.View.visualize_learning_rate(writer, lr, epoch)

visualize learning rate

tortreinador.utils.View.visualize_network(writer, net)

visualize network architecture

tortreinador.utils.View.visualize_param_hist(writer, net, epoch)

visualize histogram of params

tortreinador.utils.View.visualize_test_acc(writer, acc, epoch)

visualize test acc

tortreinador.utils.View.visualize_test_loss(writer, loss, epoch)

visualize test loss

tortreinador.utils.View.visualize_train_loss(writer, loss, n_iter)

visualize training loss

tortreinador.utils.WarmUpLR module

class tortreinador.utils.WarmUpLR.WarmUpLR(optimizer, total_iters, last_epoch=-1)

Bases: _LRScheduler

warmup_training learning rate scheduler :param optimizer: optimzier(e.g. SGD) :param total_iters: totoal_iters of warmup phase

get_lr()

we will use the first m batches, and set the learning rate to base_lr * m / total_iters

tortreinador.utils.metrics module

tortreinador.utils.metrics.mixture(pi, mu, sigma, sample_for='train')

tortreinador.utils.metrics.r2_score(y_ture, y_pred)

r2 score function for the data in GPU

Args:

• y_ture: true

• y_pred: predicted

tortreinador.utils.plot module

tortreinador.utils.plot.calculate_GMM(p, m, s, y_label)

Calculate the probability density function of the Gaussian Mixture Model

Args:

• param p: pi

• param m: mean

• param s: standard deviation

• param y_label: e.g. np.arange(0, 1, 0.001)

tortreinador.utils.plot.plot_line_2(y_1: str, y_2: str, df: DataFrame, output_path: str, fig_size: tuple = (10, 6), dpi: int = 300)

Plot Merge Line (2 Lines) using Seaborn

Args:

• param y_1: Name of Line 1

• param y_2: Name of Line 2

• param df: Dataframe

• param fig_size:

• param output_path:

• param dpi:

return: Show Line picture and save to the specific location

tortreinador.utils.preprocessing module

tortreinador.utils.preprocessing.load_data(data: DataFrame, input_parameters: list, output_parameters: list, feature_range=None, train_size: float = 0.8, val_size: float = 0.1, if_normal: bool = True, if_shuffle: bool = True, n_workers: int = 8, batch_size: int = 256, random_state=42, if_double: bool = False)

Load Data and Normalize for Regression Tasks: This function preprocesses data specifically for regression tasks by handling data splitting, optional shuffling, normalization, and DataLoader creation.

Parameters:

• data (pd.DataFrame) – The complete dataset in a Pandas DataFrame.

• input_parameters (list of str or int) – Column names or indices representing the input features.

• output_parameters (list of str or int) – Column names or indices representing the target variables.

• feature_range (tuple of (float, float), optional) – The range (min, max) used by the MinMaxScaler for scaling data. Defaults to (0, 1).

• train_size (float) – The proportion of the dataset to include in the train split (0 to 1).

• val_size (float) – The proportion of the training data to use as validation data (0 to 1).

• if_normal (bool) – Flag to determine whether to normalize the data using MinMaxScaler.

• if_shuffle (bool) – Flag to determine whether to shuffle the data before splitting into training, validation, and test sets.

• n_workers (int) – The number of subprocesses to use for data loading. More workers can increase the loading speed but consume more CPU cores.

• batch_size (int) – Number of samples per batch to load.

• random_state (int, optional) – A seed used by the random number generator for reproducibility. Defaults to None.

• if_double (bool) – Flag to determine whether to convert data to double precision (float64) format.

Returns:

Contains Train DataLoader, Validation DataLoader, Test X, Test Y, Scaler for X, and Scaler for Y. - Train DataLoader (torch.utils.data.DataLoader): DataLoader containing the training data. - Validation DataLoader (torch.utils.data.DataLoader): DataLoader containing the validation data. - Test X (np.array): Features of the test dataset. - Test Y (np.array): Targets of the test dataset. - Scaler X (sklearn.preprocessing.MinMaxScaler): Scaler object used for the input features. - Scaler Y (sklearn.preprocessing.MinMaxScaler): Scaler object used for the output targets.

Return type:

tuple

tortreinador.utils.tools module

tortreinador.utils.tools.check_outlier(validation_set, stride, windows_size, z_score_threshold)

tortreinador.utils.tools.xavier_init(net: Module)

Module contents