trixi.logger

experiment

ExperimentLogger

class trixi.logger.experiment.experimentlogger.ExperimentLogger(exp_name, base_dir, folder_format='%Y%m%d-%H%M%S_{experiment_name}', resume=False, text_logger_args=None, plot_logger_args=None, **kwargs)

Bases: trixi.logger.abstractlogger.AbstractLogger

A single class for logging your experiments to file.

It creates a experiment folder in your base folder and a folder structure to store your experiment files. The folder structure is:

base_dir/
    new_experiment_folder/
        checkpoint/
        config/
        img/
        log/
        plot/
        result/
        save/

load_checkpoint()

load_config(name, **kwargs)

Loads a config from a json file from the experiment config dir

Parameters:name – the name of the config file

Returns: A Config/ dict filled with the json file content

load_dict(path)

Loads a json file as dict from a sub path in the experiment save dir

Parameters:path – sub path to the file (starting from the experiment save dir)

Returns: The restored data as a dict

load_model()

load_numpy_data(path)

Loads a numpy file from a sub path in the experiment save dir

Parameters:path – sub path to the file (starting from the experiment save dir)

Returns: The restored numpy array

load_pickle(path)

Loads a object via pickle from a sub path in the experiment save dir

Parameters:path – sub path to the file (starting from the experiment save dir)

Returns: The restored object

resolve_format(input_, resume)

Given some input pattern, tries to find the best matching folder name by resolving the format. Options are:

• Run-number: {run_number}
• Time: “%Y%m%d-%H%M%S
• Member variables (e.g experiment_name) : {variable_name} (e.g. {experiment_name})

Parameters:

• input – The format to be resolved
• resume – Flag if folder should be resumed

Returns: The resolved folder name

save_checkpoint()

save_config(data, name, **kwargs)

Saves a config as a json file in the experiment config dir

Parameters:

• data – The data to be stored as config json
• name – The name of the json file in which the data will be stored

save_dict(data, path, indent=4, separators=(', ', ': '), encoder_cls=<class 'trixi.util.util.MultiTypeEncoder'>, **kwargs)

Saves a dict as a json file in the experiment save dir

Parameters:

• data – The data to be stored as save file
• path – sub path in the save folder (or simply filename)
• indent – Indent for the json file
• separators – Separators for the json file
• encoder_cls – Encoder Class for the encoding to json

save_file(filepath, path=None)

Copies a file to the experiment save dir

Parameters:

• filepath – Path to the file to be copied to the experiment save dir
• path – sub path to the target file (starting from the experiment save dir, does not have to exist yet)

save_model()

save_numpy_data(data, path)

Saves a numpy array in the experiment save dir

Parameters:

• data – The array to be stored as a save file
• path – sub path in the save folder (or simply filename)

save_pickle(data, path)

Saves a object data in the experiment save dir via pickle

Parameters:

• data – The data to be stored as a save file
• path – sub path in the save folder (or simply filename)

save_result(data, name, indent=4, separators=(', ', ': '), encoder_cls=<class 'trixi.util.util.MultiTypeEncoder'>, **kwargs)

Saves data as a json file in the experiment result dir

Parameters:

• data – The data to be stored as result json
• name – name of the result json file
• indent – Indent for the json file
• separators – Separators for the json file
• encoder_cls – Encoder Class for the encoding to json

show_barplot(array, name, file_format='.png', **kwargs)

This function saves a barplot in the experiment plot folder.

Parameters:

• array (np.ndarray) – array to be plotted
• name (str) – image title
• file_format (str) – file format of the image

show_boxplot(array, name, file_format='.png', **kwargs)

This function saves a boxplot in the experiment plot folder.

Parameters:

• array (np.ndarray) – array to be plotted
• name (str) – image title
• file_format (str) – file format of the image

show_image(image, name, file_format='.png', **kwargs)

This function saves an image in the experiment img folder.

Parameters:

• image (np.ndarray) – image to be shown
• name (str) – image title
• file_format (str) – file format of the image

show_lineplot(y_vals, x_vals=None, name='lineplot', file_format='.png', **kwargs)

This function saves a line plot in the experiment plot folder.

Parameters:

• x_vals – x values of the line
• y_vals – y values of the line
• name (str) – image title
• file_format (str) – file format of the image

show_matplot_plt(figure, name, file_format='.png', *args, **kwargs)

This function saves a custom matplotlib figure in the experiment plot folder.

Parameters:

• figure (matplotlib.figure.Figure) – figure to be plotted
• name (str) – image title
• file_format (str) – file format of the image

show_piechart(array, name, file_format='.png', **kwargs)

This function saves a piechart in the experiment plot folder.

Parameters:

• array (np.ndarray) – array to be plotted
• name (str) – image title
• file_format (str) – file format of the image

show_scatterplot(array, name, file_format='.png', **kwargs)

This function saves a scatterplot in the experiment plot folder.

Parameters:

• array (np.ndarray) – array to be plotted
• name (str) – image title
• file_format (str) – file format of the image

show_text(text, name=None, logger='default', **kwargs)

Logs a text to a log file.

Parameters:

• text – The text to be logged
• name – Name of the text
• logger – log file (in the experiment log folder) in which the text will be logged.
• **kwargs –

show_value(value, name=None, counter=None, tag=None, file_format='.png', **kwargs)

This function saves a value as a consequtive line plot.

Parameters:

• value (np.ndarray) – value to be plotted
• name (str) – image title
• counter – y-value of the image (if not suppled simply increases for each call)
• tag – group/label for the value. Values with the same tag will be plotted in the same plot
• file_format (str) – file format of the image

PytorchExperimentLogger

class trixi.logger.experiment.pytorchexperimentlogger.PytorchExperimentLogger(*args, **kwargs)

Bases: trixi.logger.experiment.experimentlogger.ExperimentLogger

A single class for logging your pytorch experiments to file. Extends the ExperimentLogger also also creates a experiment folder with a file structure:

The folder structure is :

base_dir/

new_experiment_folder/

checkpoint/ config/ img/ log/ plot/ result/ save/

static get_classification_metrics(tensor, labels, name='', metric=('roc-auc', 'pr-score'), use_sub_process=False, tag_name=None, results_fn=<function PytorchExperimentLogger.<lambda>>)

Displays some classification metrics as line plots in a graph (similar to show value (also uses show value for the caluclated values))

Parameters:

• tensor – Tensor with scores (e.g class probability )
• labels – Labels of the samples to which the scores match
• name – The name of the window
• metric – List of metrics to calculate. Options are: roc-auc, pr-auc, pr-score, mcc, f1
• tag_name – Name for the tag, if no given use name
• use_sub_process – Use a sub process to do the processing, if true nothing is returned
• results_fn – function which is called with the results/ return values. Expected f(val, name, tag)

Returns:

static get_input_gradient(model, inpt, err_fn, grad_type='vanilla', n_runs=20, eps=0.1, abs=False, results_fn=<function PytorchExperimentLogger.<lambda>>)

Given a model creates calculates the error and backpropagates it to the image and saves it (saliency map).

Parameters:

• model – The model to be evaluated
• inpt – Input to the model
• err_fn – The error function the evaluate the output of the model on
• grad_type – Gradient calculation method, currently supports (vanilla, vanilla-smooth, guided,
• guided-smooth) (the guided backprob can lead to segfaults -.-) –
• n_runs – Number of runs for the smooth variants
• eps – noise scaling to be applied on the input image (noise is drawn from N(0,1))
• abs (bool) – Flag, if the gradient should be a absolute value
• results_fn – function which is called with the results/ return values. Expected f(grads)

static get_pr_curve(tensor, labels, reduce_to_n_samples=None, use_sub_process=False, results_fn=<function PytorchExperimentLogger.<lambda>>)

Displays a precision recall curve given a tensor with scores and the coresponding labels

Parameters:

• tensor – Tensor with scores (e.g class probability )
• labels – Labels of the samples to which the scores match
• reduce_to_n_samples – Reduce/ downsample to to n samples for fewer data points
• use_sub_process – Use a sub process to do the processing, if true nothing is returned
• results_fn – function which is called with the results/ return values. Expected f(precision, recall)

static get_roc_curve(tensor, labels, reduce_to_n_samples=None, use_sub_process=False, results_fn=<function PytorchExperimentLogger.<lambda>>)

Displays a roc curve given a tensor with scores and the coresponding labels

Parameters:

• tensor – Tensor with scores (e.g class probability )
• labels – Labels of the samples to which the scores match
• reduce_to_n_samples – Reduce/ downsample to to n samples for fewer data points
• use_sub_process – Use a sub process to do the processing, if true nothing is returned
• results_fn – function which is called with the results/ return values. Expected f(tpr, fpr)

get_save_checkpoint_fn(name='checkpoint', **kwargs)

A function which returns a function which takes n_iter as arguments and saves the current values of the variables given as kwargs as a checkpoint file.

Parameters:

• name – Base-name of the checkpoint file
• **kwargs – dict which is actually saved, when the returned function is called

Returns: Function which takes n_iter as arguments and saves a checkpoint file

load_checkpoint(name, exclude_layer_dict=None, warnings=True, **kwargs)

Loads a checkpoint from the checkpoint directory of the experiment folder

Parameters:

• name – The name of the checkpoint file
• exclude_layer_dict – A dict with key ‘model_name’ and a list of all layers of ‘model_name’ which should
• be restored (not) –
• warnings – Flag which indicates if method should warn if not everything went perfectlys
• **kwargs – dict which is actually loaded (key=name (used to save the checkpoint) , value=variable to be
• overwritten) (loaded/) –

Returns: The kwargs dict with the loaded/ overwritten values

static load_checkpoint_static(checkpoint_file, exclude_layer_dict=None, warnings=True, **kwargs)

Loads a checkpoint/dict in a given directory (using pytorch)

Parameters:

• checkpoint_file – The checkpoint from which the checkpoint/dict should be loaded
• exclude_layer_dict – A dict with key ‘model_name’ and a list of all layers of ‘model_name’ which should
• be restored (not) –
• warnings – Flag which indicates if method should warn if not everything went perfectlys
• **kwargs – dict which is actually loaded (key=name (used to save the checkpoint) , value=variable to be
• overwritten) (loaded/) –

Returns: The kwargs dict with the loaded/ overwritten values

load_last_checkpoint(**kwargs)

Loads the (alphabetically) last checkpoint file in the checkpoint directory in the experiment folder

Parameters:

• **kwargs – dict which is actually loaded (key=name (used to save the checkpoint) , value=variable to be
• overwritten) (loaded/) –

Returns: The kwargs dict with the loaded/ overwritten values

static load_last_checkpoint_static(dir_, name=None, **kwargs)

Loads the (alphabetically) last checkpoint file in a given directory

Parameters:

• dir – The directory to look for the (alphabetically) last checkpoint
• name – String pattern which indicates the files to look form
• **kwargs – dict which is actually loaded (key=name (used to save the checkpoint) , value=variable to be
• overwritten) (loaded/) –

Returns: The kwargs dict with the loaded/ overwritten values

load_model(model, name, exclude_layers=(), warnings=True)

Loads a pytorch model from the model directory of the experiment folder

Parameters:

• model – The model to be loaded (whose parameters should be restored)
• name – The file name of the model file
• exclude_layers – List of layer names which should be excluded from restoring
• warnings – Flag which indicates if method should warn if not everything went perfectlys

static load_model_static(*args, **kwargs)

print(*args)

Prints the given arguments using the text logger print function

Parameters:*args – Things to be printed

save_at_exit(name='checkpoint_end', **kwargs)

Saves a dict as checkpoint if the program exits (not garanteed to work 100%)

Parameters:

• name – Name of the checkpoint file
• **kwargs – dict which is actually saved (key=name, value=variable to be stored)

save_checkpoint(name, n_iter=None, iter_format='{:05d}', prefix=False, **kwargs)

Saves a checkpoint in the checkpoint directory of the experiment folder

Parameters:

• name – The file name of the checkpoint file
• n_iter – The iteration number, formatted with the iter_format and added to the checkpoint name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• **kwargs – dict which is actually saved (key=name, value=variable to be stored)

static save_checkpoint_static(*args, **kwargs)

save_model(model, name, n_iter=None, iter_format='{:05d}', prefix=False)

Saves a pytorch model in the model directory of the experiment folder

Parameters:

• model – The model to be stored
• name – The file name of the model file
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix

static save_model_static(*args, **kwargs)

show_gif(frame_list=None, name='frames', scale=1.0, fps=25)

Saves gif in the img folder. Should be a list of arrays with dimension HxWxC.

Parameters:

• frame_list – The list of image tensors/arrays to be saved as a gif
• name – Filename of the gif
• scale – Scaling factor of the individual frames
• fps – FPS of the gif

show_image_gradient(name, *args, **kwargs)

Given a model creates calculates the error and backpropagates it to the image and saves it.

Parameters:

• name – Name of the file
• model – The model to be evaluated
• inpt – Input to the model
• err_fn – The error function the evaluate the output of the model on
• grad_type – Gradient calculation method, currently supports (vanilla, vanilla-smooth, guided,
• guided-smooth) (the guided backprob can lead to segfaults -.-) –
• n_runs – Number of runs for the smooth variants
• eps – noise scaling to be applied on the input image (noise is drawn from N(0,1))
• abs (bool) – Flag, if the gradient should be a absolute value

show_image_grid(image, name, **kwargs)

Saves images in the img folder as a image grid

Parameters:

• images – The images to be saved
• name – file name of the new image file

show_image_grid_heatmap(heatmap, background=None, name='heatmap', **kwargs)

Saves images in the img folder as a image grid

Parameters:

• heatmap – The images to be converted to a heatmap
• background – Context of the heatmap (to be underlayed)
• name – file name of the new image file

show_images(images, name, **kwargs)

Saves images in the img folder

Parameters:

• images – The images to be saved
• name – file name of the new image file

show_video(frame_list=None, name='video', dim='LxHxWxC', scale=1.0, fps=25, extension='.mp4', codec='THEO')

Saves video in the img folder. Should be a list of arrays with dimension HxWxC.

Parameters:

• frame_list – The list of image tensors/arrays to be saved as a video
• name – Filename of the video
• dim – Dimension of the tensor - should be either LxHxWxC or LxCxHxW
• fps – FPS of the video
• extension – File extension - should be mp4, ogc, avi or webm

file

NumpyPlotFileLogger

class trixi.logger.file.numpyplotfilelogger.NumpyPlotFileLogger(img_dir, plot_dir, switch_backend=True, **kwargs)

Bases: trixi.logger.plt.numpyseabornplotlogger.NumpySeabornPlotLogger

NumpyPlotFileLogger is a logger, which can plot/ interpret numpy array as different types (images, lineplots, …) into an image and plot directory. For the plotting it builds up on the NumpySeabornPlotLogger.

show_barplot(array, name, file_format='.png', *args, **kwargs)

Method which creates and stores a barplot

Parameters:

• array – Array of values you want to plot
• name – file-name
• file_format – output-image (plot) file format

show_boxplot(array, name, file_format='.png', *args, **kwargs)

Method which creates and stores a boxplot

Parameters:

• array – Array of values you want to plot
• name – file-name
• file_format – output-image (plot) file format

show_image(image, name, file_format='.png', *args, **kwargs)

Method which stores an image as a image file

Parameters:

• image – Numpy array-image
• name – file-name
• file_format – output-image file format

show_lineplot(y_vals, x_vals, name, file_format='.png', *args, **kwargs)

Method which creates and stores a lineplot

Parameters:

• y_vals – Array of y values
• x_vals – Array of corresponding x-values
• name – file-name
• file_format – output-image (plot) file format

show_matplot_plt(figure, name, file_format='.png', *args, **kwargs)

Method to save a custom matplotlib figure

Parameters:

• figure – Figure you want to plot
• name – file name
• file_format – output image (plot) file format

show_piechart(array, name, file_format='.png', *args, **kwargs)

Method which creates and stores a piechart

Parameters:

• array – Array of values you want to plot
• name – file-name
• file_format – output-image (plot) file format

show_scatterplot(array, name, file_format='.png', *args, **kwargs)

Method which creates and stores a scatter

Parameters:

• array – Array of values you want to plot
• name – file-name
• file_format – output-image (plot) file format

show_value(value, name, counter=None, tag=None, file_format='.png', *args, **kwargs)

Method which logs a value as a line plot

Parameters:

• value – Value (y-axis value) you want to display/ plot/ store
• name – Name of the value (will also be the filename if no tag is given)
• counter – counter, which tells the number of the sample (with the same name –> filename) (x-axis value)
• tag – Tag, grouping similar values. Values with the same tag will be plotted in the same plot
• file_format – output-image file format

Returns:

trixi.logger.file.numpyplotfilelogger.threaded(func)

PytorchPlotFileLogger

class trixi.logger.file.pytorchplotfilelogger.PytorchPlotFileLogger(*args, **kwargs)

Bases: trixi.logger.file.numpyplotfilelogger.NumpyPlotFileLogger

Visual logger, inherits the NumpyPlotLogger and plots/ logs pytorch tensors and variables as files on the local file system.

process_params(f, *args, **kwargs)

Inherited “decorator”: convert Pytorch variables and Tensors to numpy arrays

save_image(tensor, name, n_iter=None, iter_format='{:05d}', prefix=False, image_args=None)

Saves an image into the image directory of the PytorchPlotFileLogger

Parameters:

• tensor – Tensor containing the image
• name – file-name of the image file
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

save_image_grid(tensor, name, n_iter=None, prefix=False, iter_format='{:05d}', image_args=None)

Saves images of a 4d- tensor (N, C, H, W) as a image grid into an image file in the image directory of the PytorchPlotFileLogger

Parameters:

• tensor – 4d- tensor (N, C, H, W)
• name – file-name of the image file
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

static save_image_grid_static(*args, **kwargs)

static save_image_static(*args, **kwargs)

save_images(tensors, n_iter=None, iter_format='{:05d}', prefix=False, image_args=None)

Saves an image tensors into the image directory of the PytorchPlotFileLogger

Parameters:

• tensors – A dict with file-name-> tensor to plot as image
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

static save_images_static(*args, **kwargs)

show_image(image, name, n_iter=None, iter_format='{:05d}', prefix=False, image_args=None, **kwargs)

Calls the save image method (for abstract logger combatibility)

Parameters:

• image – Tensor containing the image
• name – file-name of the image file
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

show_image_grid(images, name, n_iter=None, prefix=False, iter_format='{:05d}', image_args=None, **kwargs)

Calls the save image grid method (for abstract logger combatibility)

Parameters:

• images – 4d- tensor (N, C, H, W)
• name – file-name of the image file
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

show_image_grid_heatmap(heatmap, background=None, ratio=0.3, normalize=True, colormap=<sphinx.ext.autodoc.importer._MockObject object>, name='heatmap', n_iter=None, prefix=False, iter_format='{:05d}', image_args=None, **kwargs)

Creates heat map from the given map and if given combines it with the background and then displays results with as image grid.

Parameters:

• heatmap – 4d- tensor (N, C, H, W) to be converted to a heatmap
• background – 4d- tensor (N, C, H, W) background/ context of the heatmap (to be underlayed)
• name – The name of the window
• ratio – The ratio to mix the map with the background (0 = only background, 1 = only map)
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

show_images(images, name, n_iter=None, iter_format='{:05d}', prefix=False, image_args=None, **kwargs)

Calls the save images method (for abstract logger combatibility)

Parameters:

• images – List of Tensors
• name – List of file names (corresponding to the images list)
• n_iter – The iteration number, formatted with the iter_format and added to the model name (if not None)
• iter_format – The format string, which indicates how n_iter will be formated as a string
• prefix – If True, the formated n_iter will be appended as a prefix, otherwise as a suffix
• image_args – Arguments for the tensorvision save image method

TextFileLogger

class trixi.logger.file.textfilelogger.TextFileLogger(base_dir=None, logging_level=10, logging_stream=<_io.TextIOWrapper name='<stdout>' mode='w' encoding='UTF-8'>, default_stream_handler=True, **kwargs)

Bases: trixi.logger.abstractlogger.AbstractLogger

A Logger for logging text into different text files and output streams (using the python logging framework)

add_file_handler(name, logger='default')

Adds a file handler to a logger, thus the logger will log into a log file with a given name

Parameters:

• name – File name of the log file (in which the logger will log now)
• logger – Name of the logger to add the file-hander/ logging file to

add_handler(handler, logger='default')

Adds an additional handler to a logger with a name :param handler: Logging handler to be added to a given logger :param logger: Name of the logger to add the hander to

add_logger(name, logging_level=None, file_handler=True, stream_handler=True)

Adds a new logger

Parameters:

• name – Name of the new logger
• logging_level – Logging level of the new logger
• file_handler – Flag, if it should use a file_handler, if yes creates a new file with the given name in the logging directory
• stream_handler – Flag, if the logger should also log to the default stream

Returns:

add_stream_handler(logger='default')

Adds a stream handler to a logger, thus the logger will log into the default logging stream

Parameters:logger – Name of the logger to add the stream-hander to

debug(msg, logger='default')

Prints and logs a message with the level debug

Parameters:

• msg – Message to print/ log
• logger – Logger which should log

error(msg, logger='default')

Prints and logs a message with the level error

Parameters:

• msg – Message to print/ log
• logger – Logger which should log

info(msg, logger='default')

Prints and logs a message with the level info

Parameters:

• msg – Message to print/ log
• logger – Logger which should log

log(msg, logger='default')

Prints and logs a message with the level info

Parameters:

• msg – Message to print/ log
• logger – Logger which should log

log_to(msg, name, log_to_default=False)

Logs to an existing logger or if logger does not exists creates new one

Parameters:

• msg – Message to be logged
• name – Name of the logger to log to (usually also the logfile-name)
• log_to_default – Flag if it should in addition to the logger given by name, log to the default logger

print(*args, logger='default')

Prints and logs objects

Parameters:

• *args – Object to print/ log
• logger – Logger which should log

show_text(text, name=None, logger='default', **kwargs)

Logs a text. Calls the log function (for combatibility reasons with AbstractLogger)

Parameters:

• text – Text to be logged
• name – Some identifier for the text (will be added infront of the text)
• logger – Name of the Logger to log to

show_value(value, name=None, logger='default', **kwargs)

Logs a Value. Calls the log function (for combatibility reasons with AbstractLogger)

Parameters:

• value – Value to be logged
• name – Some identifier for the text (will be added infront of the text)
• logger – Name of the Logger to log to

message

SlackMessageLogger

class trixi.logger.message.slackmessagelogger.SlackMessageLogger(token, user_email, exp_name=None, *args, **kwargs)

Bases: trixi.logger.plt.numpyseabornimageplotlogger.NumpySeabornImagePlotLogger

Slack logger, inherits the NumpySeabornImagePlotLogger and sends plots/logs to a chat via a Slack bot.

delete_message(ts)

Deletes a direct message from the bot with the given timestamp (ts)

Parameters:ts – Time stamp the message was sent

static find_cid_for_user(slack_client, uid)

Returns the chat/channel id for a direct message of the bot with the given user

Parameters:

• slack_client – Slack client (already authorized)
• uid – User id of the user

Returns:

chat/channel id for a direct message

static find_uid_for_email(slack_client, email)

Returns the slack user id for a given email

Parameters:

• slack_client – Slack client (already authorized)
• email – Workspace email address to get the user id for

Returns:

Slack workspace user id

print(text, *args, **kwargs)

Just calls show_text()

process_params(f, *args, **kwargs)

Inherited “decorator”: convert PyTorch variables and Tensors to numpy arrays

send_message(message='', file=None)

Sends a message and a file if one is given

Parameters:

• message – Message to be sent
• file – File to be sent

Returns:

The timestamp (ts) of the message

show_barplot(array, name='barplot', delete_last=True, *args, **kwargs)

Sends a barplot to a chat using an existing slack bot.

Parameters:

• array – array of shape NxM where N is the number of rows and M is the number of elements in the row.
• name – The name of the figure
• delete_last – If a message with the same name was sent, delete it beforehand

show_image(image, *args, **kwargs)

Sends an image file to a chat using an existing slack bot.

Parameters:image (str or np array) – Path to the image file to be sent to the chat.

show_image_grid(image_array, name=None, nrow=8, padding=2, normalize=False, range=None, scale_each=False, pad_value=0, delete_last=True, *args, **kwargs)

Sends an array of images to a chat using an existing Slack bot. (Requires torch and torchvision)

Parameters:

• image_array (np.narray / torch.tensor) – Image array/tensor which will be sent as an image grid
• make_grid_kargs – Key word arguments for the torchvision make grid method
• delete_last – If a message with the same name was sent, delete it beforehand

show_lineplot(y_vals, x_vals=None, name='lineplot', delete_last=True, *args, **kwargs)

Sends a lineplot to a chat using an existing slack bot.

Parameters:

• y_vals – Array of shape MxN , where M is the number of points and N is the number of different line
• x_vals – Has to have the same shape as Y: MxN. For each point in Y it gives the corresponding X value (if not set the points are assumed to be equally distributed in the interval [0, 1])
• name – The name of the figure
• delete_last – If a message with the same name was sent, delete it beforehand

show_piechart(array, name='piechart', delete_last=True, *args, **kwargs)

Sends a piechart to a chat using an existing slack bot.

Parameters:

• array – Array of positive integers. Each integer will be presented as a part of the pie (with the total as the sum of all integers)
• name – The name of the figure
• delete_last – If a message with the same name was sent, delete it beforehand

show_scatterplot(array, name='scatterplot', delete_last=True, *args, **kwargs)

Sends a scatterplot to a chat using an existing slack bot.

Parameters:

• array – An array with size N x dim, where each element i in N at X[i] results in a 2D (if dim = 2) or 3D (if dim = 3) point.
• name – The name of the figure
• delete_last – If a message with the same name was sent, delete it beforehand

show_text(text, *args, **kwargs)

Sends a text to a chat using an existing slack bot.

Parameters:text (str) – Text message to be sent to the bot.

show_value(value, name, counter=None, tag=None, delete_last=True, *args, **kwargs)

Sends a value to a chat using an existing slack bot.

Parameters:

• value – Value to be plotted sent to the chat.
• name – Name for the plot.
• counter – Optional counter to be sent in conjunction with the value.
• tag – Tag to be used as a label for the plot.
• delete_last – If a message with the same name was sent, delete it beforehand

TelegramMessageLogger

class trixi.logger.message.telegrammessagelogger.TelegramMessageLogger(token, chat_id, exp_name=None, *args, **kwargs)

Bases: trixi.logger.plt.numpyseabornimageplotlogger.NumpySeabornImagePlotLogger

Telegram logger, inherits the NumpySeabornImagePlotLogger and sends plots/logs to a chat via a Telegram bot.

print(text, *args, **kwargs)

Just calls show_text()

process_params(f, *args, **kwargs)

Inherited “decorator”: convert PyTorch variables and Tensors to numpy arrays

show_barplot(array, name=None, *args, **kwargs)

Sends a barplot to a chat using an existing Telegram bot.

Parameters:

• array – array of shape NxM where N is the number of rows and M is the number of elements in the row.
• name – The name of the figure

show_image(image, *args, **kwargs)

Sends an image file to a chat using an existing Telegram bot.

Parameters:image (str or np array) – Path to the image file to be sent to the chat.

show_image_grid(image_array, name=None, nrow=8, padding=2, normalize=False, range=None, scale_each=False, pad_value=0, *args, **kwargs)

Sends an array of images to a chat using an existing Telegram bot. (Requires torch and torchvision)

Parameters:

• image_array (np.narray / torch.tensor) – Image array/ tensor which will be sent as an image grid
• make_grid_kargs – Key word arguments for the torchvision make grid method

show_lineplot(y_vals, x_vals=None, name=None, *args, **kwargs)

Sends a lineplot to a chat using an existing Telegram bot.

Parameters:

• y_vals – Array of shape MxN , where M is the number of points and N is the number of different line
• x_vals – Has to have the same shape as Y: MxN. For each point in Y it gives the corresponding X value (if
• set the points are assumed to be equally distributed in the interval [0, 1] ) (not) –
• name – The name of the figure

show_piechart(array, name=None, *args, **kwargs)

Sends a piechart to a chat using an existing Telegram bot.

Parameters:

• array – Array of positive integers. Each integer will be presented as a part of the pie (with the total
• the sum of all integers) (as) –
• name – The name of the figure

show_scatterplot(array, name=None, *args, **kwargs)

Sends a scatterplot to a chat using an existing Telegram bot.

Parameters:

• array – A 2d array with size N x dim, where each element i in N at X[i] results in a a 2d (if dim = 2)/
• 3d (if dim = 3) –
• name – The name of the figure

show_text(text, *args, **kwargs)

Sends a text to a chat using an existing Telegram bot.

Parameters:text (str) – Text message to be sent to the bot.

show_value(value, name, counter=None, tag=None, *args, **kwargs)

Sends a value to a chat using an existing Telegram bot.

Parameters:

• value – Value to be plotted sent to the chat.
• name – Name for the plot.
• counter – Optional counter to be sent in conjunction with the value.
• tag – Tag to be used as a label for the plot.

plt

NumpySeabornPlotLogger

class trixi.logger.plt.numpyseabornplotlogger.NumpySeabornPlotLogger(**kwargs)

Bases: trixi.logger.abstractlogger.AbstractLogger

Visual logger, inherits the AbstractLogger and plots/ logs numpy arrays/ values as matplotlib / seaborn plots.

get_figure(name)

Returns a figure with a given name as identifier.

If no figure yet exists with the name a new one is created. Otherwise the existing one is returned

Parameters:name – Name of the figure Returns:A figure with the given name

show_barplot(array, name=None, show=True, *args, **kwargs)

Creates a bar plot figure from an array

Parameters:

• array – array of shape NxM where N is the number of rows and M is the number of elements in the row.
• name – The name of the figure
• show – Flag if it should also display the figure (result might also depend on the matplotlib backend )

Returns:

A matplotlib figure

show_boxplot(array, name, show=True, *args, **kwargs)

Creates a box plot figure from an array

Parameters:

• array – array of shape NxM where N is the number of rows and M is the number of elements in the row.
• name – The name of the figure
• show – Flag if it should also display the figure (result might also depend on the matplotlib backend )

Returns:

A matplotlib figure

show_image(image, name=None, show=True, *args, **kwargs)

Create an image figure

Parameters:

• image – The image array to be displayed
• name – The name of the image window
• show – Flag if it should also display the figure (result might also depend on the matplotlib backend )

Returns:

A matplotlib figure

show_lineplot(y_vals, x_vals=None, name=None, show=True, *args, **kwargs)

Creates a line plot figure with (multiple) lines plot, given values Y (and optional the corresponding X values)

Parameters:

• y_vals – Array of shape MxN , where M is the number of points and N is the number of different line
• x_vals – Has to have the same shape as Y: MxN. For each point in Y it gives the corresponding X value (if
• set the points are assumed to be equally distributed in the interval [0, 1] ) (not) –
• name – The name of the figure
• show – Flag if it should also display the figure (result might also depend on the matplotlib backend )

Returns:

A matplotlib figure

show_piechart(array, name=None, show=True, *args, **kwargs)

Creates a scatter plot figure

Parameters:

• array – Array of positive integers. Each integer will be presented as a part of the pie (with the total
• the sum of all integers) (as) –
• name – The name of the figure
• show – Flag if it should also display the figure (result might also depend on the matplotlib backend )

Returns:

A matplotlib figure

show_scatterplot(array, name=None, show=True, *args, **kwargs)

Creates a scatter plot figure with the points given in array

Parameters:

• array – A 2d array with size N x dim, where each element i in N at X[i] results in a a 2d (if dim = 2)/
• 3d (if dim = 3) –
• name – The name of the figure
• show – Flag if it should also display the figure (result might also depend on the matplotlib backend )

Returns:

A matplotlib figure

show_value(value, name, counter=None, tag=None, show=True, *args, **kwargs)

Creates a line plot that is automatically appended with new values and returns it as a figure.

Parameters:

• value – Value to be plotted / appended to the graph (y-axis value)
• name – The name of the window
• counter – counter, which tells the number of the sample (with the same name) (x-axis value)
• tag – Tag, grouping similar values. Values with the same tag will be plotted in the same plot
• show –

  Flag if it should also display the figure (result might also depend on the matplotlib backend )

  Returns:

  A matplotlib figure

NumpySeabornImagePlotLogger

class trixi.logger.plt.numpyseabornimageplotlogger.NumpySeabornImagePlotLogger(**kwargs)

Bases: trixi.logger.plt.numpyseabornplotlogger.NumpySeabornPlotLogger

Wrapper around NumpySeabornPlotLogger that renders figures into numpy arrays.

show_barplot(array, name=None, *args, **kwargs)

Creates a bar plot figure from an array

Parameters:

• array – array of shape NxM where N is the number of rows and M is the number of elements in the row.
• name – The name of the figure

Returns:

A numpy array image of the figure

show_image(image, name=None, *args, **kwargs)

Create an image figure

Parameters:

• image – The image array to be displayed
• name – The name of the image window

Returns:

A numpy array image of the figure

show_lineplot(y_vals, x_vals=None, name=None, *args, **kwargs)

Creates a line plot figure with (multiple) lines plot, given values Y (and optional the corresponding X values)

Parameters:

• y_vals – Array of shape MxN , where M is the number of points and N is the number of different line
• x_vals – Has to have the same shape as Y: MxN. For each point in Y it gives the corresponding X value (if
• set the points are assumed to be equally distributed in the interval [0, 1] ) (not) –
• name – The name of the figure

Returns:

A numpy array image of the figure

show_piechart(array, name=None, *args, **kwargs)

Creates a scatter plot figure

Parameters:

• array – Array of positive integers. Each integer will be presented as a part of the pie (with the total
• the sum of all integers) (as) –
• name – The name of the figure

Returns:

A numpy array image of the figure

show_scatterplot(array, name=None, *args, **kwargs)

Creates a scatter plot figure with the points given in array

Parameters:

• array – A 2d array with size N x dim, where each element i in N at X[i] results in a a 2d (if dim = 2)/
• 3d (if dim = 3) –
• name –

  The name of the figure

  Returns:

  A numpy array image of the figure

show_value(value, name, counter=None, tag=None, *args, **kwargs)

Creates a line plot that is automatically appended with new values and returns it as a figure.

Parameters:

• value – Value to be plotted / appended to the graph (y-axis value)
• name – The name of the window
• counter – counter, which tells the number of the sample (with the same name) (x-axis value)
• tag –

  Tag, grouping similar values. Values with the same tag will be plotted in the same plot

  Returns:

  A numpy array image of the figure

tensorboard

TensorboardLogger

PytorchTensorboardLogger

visdom

NumpyVisdomLogger

class trixi.logger.visdom.numpyvisdomlogger.NumpyVisdomLogger(exp_name='main', server='http://localhost', port=8080, auto_close=True, auto_start=False, auto_start_ports=(8080, 8000), **kwargs)

Bases: trixi.logger.abstractlogger.AbstractLogger

Visual logger, inherits the AbstractLogger and plots/ logs numpy arrays/ values on a Visdom server.

add_to_graph(*args, **kwargs)

close_all()

Closes all visdom windows.

exit()

Kills the internal process.

save_vis()

send_data(*args, **kwargs)

show_barplot(*args, **kwargs)

show_boxplot(*args, **kwargs)

show_contourplot(*args, **kwargs)

show_histogram(*args, **kwargs)

show_image(*args, **kwargs)

show_images(*args, **kwargs)

show_lineplot(*args, **kwargs)

show_matplot_plt(*args, **kwargs)

show_piechart(*args, **kwargs)

show_plotly_plt(*args, **kwargs)

show_progress(*args, **kwargs)

show_scatterplot(*args, **kwargs)

show_surfaceplot(*args, **kwargs)

show_svg(*args, **kwargs)

show_text(*args, **kwargs)

show_value(*args, **kwargs)

show_values(val_dict)

A util function for multiple values. Simple plots all values in a dict, there the window name is the key in the dict and the plotted value is the value of a dict (Simply calls the show_value function).

Parameters:val_dict – Dict with key, values pairs which will be plotted

trixi.logger.visdom.numpyvisdomlogger.add_to_queue(func)

trixi.logger.visdom.numpyvisdomlogger.start_visdom(port_list=(8080, 8000))

Starts a visdom server on a given port

Parameters:port_list – Priority list of port. Will start a visdom server on the first available port.

PytorchVisdomLogger

class trixi.logger.visdom.pytorchvisdomlogger.PytorchVisdomLogger(*args, **kwargs)

Bases: trixi.logger.visdom.numpyvisdomlogger.NumpyVisdomLogger

Visual logger, inherits the NumpyVisdomLogger and plots/ logs pytorch tensors and variables on a Visdom server.

plot_model_gradient_flow(model, name='model', title=None)

Plots statstics (mean, std, abs(max)) of the weights or the corresponding gradients of a model as a barplot.

Parameters:

• model – Model with the weights.
• env_appendix – Visdom environment name appendix, if none is given, it uses “-histogram”.
• model_name – Name of the model (is used as window name).
• plot_grad – If false plots weight statistics, if true plot the gradients of the weights.

plot_model_statistics(**kwargs)

plot_model_statistics_grads(model, env_appendix='', model_name='', **kwargs)

Plots statstics (mean, std, abs(max)) of the gradients of a model as a barplot (uses plot model statistics with plot_grad=True).

Parameters:

• model – Model with the weights and the corresponding gradients (have to calculated previously).
• env_appendix – Visdom environment name appendix
• model_name – Name of the model (is used as window name).

plot_model_statistics_weights(model, env_appendix='', model_name='', **kwargs)

Plots statstics (mean, std, abs(max)) of the weights of a model as a barplot (uses plot model statistics with plot_grad=False).

Parameters:

• model – Model with the weights.
• env_appendix – Visdom environment name appendix
• model_name – Name of the model (is used as window name).

plot_model_structure(model, input_size, name='model_structure', use_cuda=True, delete_tmp_on_close=False, forward_kwargs=None, **kwargs)

Plots the model structure/ model graph of a pytorch module (this only works correctly with pytorch 0.2.0).

Parameters:

• model – The graph of this model will be plotted.
• input_size – Input size of the model (with batch dim).
• name – The name of the window in the visdom env.
• use_cuda – Perform model dimension calculations on the gpu (cuda).
• delete_tmp_on_close – Determines if the tmp file will be deleted on close. If set true, can cause problems due to the multi threadded plotting.

plot_mutliple_models_statistics_grads(model_dict, env_appendix='', **kwargs)

Given models in a dict, plots the gradient statistics of the models.

Parameters:

• model_dict – Dict with models, the key is assumed to be the name, while the value is the model.
• env_appendix – Visdom environment name appendix

plot_mutliple_models_statistics_weights(model_dict, env_appendix=None, **kwargs)

Given models in a dict, plots the weight statistics of the models.

Parameters:

• model_dict – Dict with models, the key is assumed to be the name, while the value is the model.
• env_appendix – Visdom environment name appendix

process_params(f, *args, **kwargs)

Inherited “decorator”: convert Pytorch variables and Tensors to numpy arrays.

show_classification_metrics(tensor, labels, name, metric=('roc-auc', 'pr-score'), use_sub_process=False, tag_name=None)

Displays some classification metrics as line plots in a graph (similar to show value (also uses show value for the caluclated values))

Parameters:

• tensor – Tensor with scores (e.g class probability )
• labels – Labels of the samples to which the scores match
• name – The name of the window
• metric – List of metrics to calculate. Options are: roc-auc, pr-auc, pr-score, mcc, f1

Returns:

show_embedding(tensor, labels=None, name=None, method='tsne', n_dims=2, n_neigh=30, meth_args=None, *args, **kwargs)

Displays a tensor a an embedding

Parameters:

• tensor – Tensor to be embedded an then displayed
• labels – Labels of the entries in the tensor (first dimension)
• name – The name of the window
• method – Method used for embedding, options are: tsne, standard, ltsa, hessian, modified, isomap, mds,
• umap (spectral,) –
• n_dims – dimensions to embed the data into
• n_neigh – Neighbour parameter to kind of determin the embedding (see t-SNE for more information)
• meth_args – Further arguments which can be passed to the embedding method

show_image_gradient(model, inpt, err_fn, grad_type='vanilla', n_runs=20, eps=0.1, abs=False, **image_grid_params)

Given a model creates calculates the error and backpropagates it to the image and saves it (saliency map).

Parameters:

• model – The model to be evaluated
• inpt – Input to the model
• err_fn – The error function the evaluate the output of the model on
• grad_type – Gradient calculation method, currently supports (vanilla, vanilla-smooth, guided,
• guided-smooth) (the guided backprob can lead to segfaults -.-) –
• n_runs – Number of runs for the smooth variants
• eps – noise scaling to be applied on the input image (noise is drawn from N(0,1))
• abs (bool) – Flag, if the gradient should be a absolute value
• **image_grid_params – Params for make image grid.

show_image_grid(**kwargs)

show_image_grid_heatmap(*args, **kwargs)

show_pr_curve(tensor, labels, name, reduce_to_n_samples=None, use_sub_process=False)

Displays a precision recall curve given a tensor with scores and the coresponding labels

Parameters:

• tensor – Tensor with scores (e.g class probability )
• labels – Labels of the samples to which the scores match
• name – The name of the window
• reduce_to_n_samples – Reduce/ downsample to to n samples for fewer data points
• use_sub_process – Use a sub process to do the processing

show_roc_curve(tensor, labels, name, reduce_to_n_samples=None, use_sub_process=False)

Displays a roc curve given a tensor with scores and the coresponding labels

Parameters:

• tensor – Tensor with scores (e.g class probability )
• labels – Labels of the samples to which the scores match
• name – The name of the window
• reduce_to_n_samples – Reduce/ downsample to to n samples for fewer data points
• use_sub_process – Use a sub process to do the processing

show_video(frame_list=None, name='frames', dim='LxHxWxC', scale=1.0, fps=25)

trixi.logger.visdom.pytorchvisdomlogger.move_to_cpu(fn)

Decorator to call the process_params method of the class.

AbstractLogger

class trixi.logger.abstractlogger.AbstractLogger(*args, **kwargs)

Bases: object

Abstract interface for visual logger.

process_params(f, *args, **kwargs)

Implement this to handle data conversions in your logger.

Example: Implement logger for numpy data, then implement torch logger as child of numpy logger and just use the process_params method to convert from torch to numpy.

show_barplot(*args, **kwargs)

Abstract method which should handle and somehow log/ store a barplot

show_image(*args, **kwargs)

Abstract method which should handle and somehow log/ store an image

show_lineplot(*args, **kwargs)

Abstract method which should handle and somehow log/ store a lineplot

show_piechart(*args, **kwargs)

Abstract method which should handle and somehow log/ store a piechart

show_scatterplot(*args, **kwargs)

Abstract method which should handle and somehow log/ store a scatterplot

show_text(*args, **kwargs)

Abstract method which should handle and somehow log/ store a text

show_value(*args, **kwargs)

Abstract method which should handle and somehow log/ store a value

trixi.logger.abstractlogger.convert_params(f)

Decorator to call the process_params method of the class.

trixi.logger.abstractlogger.threaded(f)

Decorator to run the process in an extra thread.

CombinedLogger

class trixi.logger.combinedlogger.CombinedLogger(*loggers)

Bases: object

A Logger which can combine all other logger and if called calls all the sub loggers

trixi.logger.combinedlogger.create_function(self, sub_methods)