Module ainshamsflow.utils.history
History Object Module.
In this Module, we define the History Class used to visualize statistics such as loss and metric values during training.
Expand source code
"""History Object Module.
In this Module, we define the History Class used to visualize statistics
such as loss and metric values during training.
"""
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.ticker import MaxNLocator
class History:
"""History Class."""
def __init__(self, loss, metrics=None, valid=False):
"""Initialize empty loss and metrics values and save loss and metrics names."""
if metrics is None:
metrics = []
self.valid = valid
self.loss_values = []
self.metric_values = []
if valid:
self.val_loss_values = []
self.val_metric_values = []
self.loss_name = loss.__name__
self.metric_names = [metric.__name__ for metric in metrics]
if len(self.loss_name) > 12:
if self.loss_name[-12:] == 'Crossentropy':
self.loss_name = 'Crossentropy'
def add(self, loss_value, metric_values, val_loss_value=None, val_metric_values=None):
"""Add Values to the history of the training session."""
self.loss_values.append(loss_value)
self.metric_values.append(metric_values)
if self.valid:
self.val_loss_values.append(val_loss_value)
self.val_metric_values.append(val_metric_values)
def flipped_metrics(self):
"""Return the metric Values as flipped for easy visualization and data extraction."""
return np.array(self.metric_values).T
def flipped_valid_metrics(self):
"""Return the metric Values as flipped for easy visualization and data extraction."""
assert self.valid
return np.concatenate([
np.array(self.metric_values).T,
np.array(self.val_metric_values).T
], axis=0)
def show(self, show_metrics=True):
"""Show the training statistics in an image from saved data."""
if self.valid:
num_of_plots = len(self.metric_names) + 1 if show_metrics else 1
num_of_plots *= 2
fig, axs = plt.subplots(num_of_plots, 1, squeeze=False, figsize=(8, 3 * num_of_plots))
fig.patch.set_facecolor('w')
# Loss Plots:
axs[0, 0].plot(self.loss_values)
axs[0, 0].set_title('Model Loss')
axs[0, 0].set_ylabel('loss')
bottom, top = axs[0, 0].get_ylim()
axs[0, 0].set_ylim(0, top)
axs[0, 0].xaxis.set_major_locator(MaxNLocator(integer=True))
axs[1, 0].plot(self.val_loss_values)
axs[1, 0].set_ylabel('val_loss')
axs[1, 0].set_xlabel('epochs')
bottom, top = axs[1, 0].get_ylim()
axs[1, 0].set_ylim(0, top)
axs[1, 0].xaxis.set_major_locator(MaxNLocator(integer=True))
# Metrics Plots:
if self.metric_names and show_metrics:
for j, (metric_name, metric_values, valid_values) in enumerate(zip(self.metric_names, self.flipped_metrics(), self.flipped_valid_metrics())):
if j == 0:
axs[2*j + 2, 0].set_title('Model Metrics')
axs[2*j + 2, 0].plot(metric_values)
axs[2*j + 3, 0].plot(valid_values)
axs[2*j + 2, 0].set_ylabel(metric_name)
axs[2*j + 3, 0].set_ylabel('val_' + metric_name)
if len(metric_name) == 2:
bottom, top = axs[2*j + 2, 0].get_ylim()
axs[2*j + 2, 0].set_ylim(0, top)
bottom, top = axs[2*j + 3, 0].get_ylim()
axs[2*j + 3, 0].set_ylim(0, top)
else:
axs[2*j + 2, 0].set_ylim(0, 1)
axs[2*j + 3, 0].set_ylim(0, 1)
axs[2*j + 2, 0].xaxis.set_major_locator(MaxNLocator(integer=True))
axs[2*j + 3, 0].xaxis.set_major_locator(MaxNLocator(integer=True))
axs[-1, 0].set_xlabel('epochs')
else:
num_of_plots = len(self.metric_names) + 1 if show_metrics else 1
fig, axs = plt.subplots(num_of_plots, 1, squeeze=False, figsize=(8, 3*num_of_plots))
fig.patch.set_facecolor('w')
# Loss Plots:
axs[0, 0].plot(self.loss_values)
axs[0, 0].set_title('Model Loss')
axs[0, 0].set_ylabel('loss')
axs[0, 0].set_xlabel('epochs')
bottom, top = axs[0, 0].get_ylim()
axs[0, 0].set_ylim(0, top)
axs[0, 0].xaxis.set_major_locator(MaxNLocator(integer=True))
# Metrics Plots:
if self.metric_names and show_metrics:
for j, metric_values in enumerate(self.flipped_metrics()):
axs[j+1, 0].plot(metric_values)
if j == 0:
axs[j+1, 0].set_title('Model Metrics')
axs[j+1, 0].set_ylabel(self.metric_names[j])
if len(self.metric_names[j]) == 2:
bottom, top = axs[j+1, 0].get_ylim()
axs[j+1, 0].set_ylim(0, top)
else:
axs[j+1, 0].set_ylim(0, 1)
axs[j+1, 0].xaxis.set_major_locator(MaxNLocator(integer=True))
axs[-1, 0].set_xlabel('epochs')
plt.show()Classes
class History (loss, metrics=None, valid=False)-
History Class.
Initialize empty loss and metrics values and save loss and metrics names.
Expand source code
class History: """History Class.""" def __init__(self, loss, metrics=None, valid=False): """Initialize empty loss and metrics values and save loss and metrics names.""" if metrics is None: metrics = [] self.valid = valid self.loss_values = [] self.metric_values = [] if valid: self.val_loss_values = [] self.val_metric_values = [] self.loss_name = loss.__name__ self.metric_names = [metric.__name__ for metric in metrics] if len(self.loss_name) > 12: if self.loss_name[-12:] == 'Crossentropy': self.loss_name = 'Crossentropy' def add(self, loss_value, metric_values, val_loss_value=None, val_metric_values=None): """Add Values to the history of the training session.""" self.loss_values.append(loss_value) self.metric_values.append(metric_values) if self.valid: self.val_loss_values.append(val_loss_value) self.val_metric_values.append(val_metric_values) def flipped_metrics(self): """Return the metric Values as flipped for easy visualization and data extraction.""" return np.array(self.metric_values).T def flipped_valid_metrics(self): """Return the metric Values as flipped for easy visualization and data extraction.""" assert self.valid return np.concatenate([ np.array(self.metric_values).T, np.array(self.val_metric_values).T ], axis=0) def show(self, show_metrics=True): """Show the training statistics in an image from saved data.""" if self.valid: num_of_plots = len(self.metric_names) + 1 if show_metrics else 1 num_of_plots *= 2 fig, axs = plt.subplots(num_of_plots, 1, squeeze=False, figsize=(8, 3 * num_of_plots)) fig.patch.set_facecolor('w') # Loss Plots: axs[0, 0].plot(self.loss_values) axs[0, 0].set_title('Model Loss') axs[0, 0].set_ylabel('loss') bottom, top = axs[0, 0].get_ylim() axs[0, 0].set_ylim(0, top) axs[0, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[1, 0].plot(self.val_loss_values) axs[1, 0].set_ylabel('val_loss') axs[1, 0].set_xlabel('epochs') bottom, top = axs[1, 0].get_ylim() axs[1, 0].set_ylim(0, top) axs[1, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) # Metrics Plots: if self.metric_names and show_metrics: for j, (metric_name, metric_values, valid_values) in enumerate(zip(self.metric_names, self.flipped_metrics(), self.flipped_valid_metrics())): if j == 0: axs[2*j + 2, 0].set_title('Model Metrics') axs[2*j + 2, 0].plot(metric_values) axs[2*j + 3, 0].plot(valid_values) axs[2*j + 2, 0].set_ylabel(metric_name) axs[2*j + 3, 0].set_ylabel('val_' + metric_name) if len(metric_name) == 2: bottom, top = axs[2*j + 2, 0].get_ylim() axs[2*j + 2, 0].set_ylim(0, top) bottom, top = axs[2*j + 3, 0].get_ylim() axs[2*j + 3, 0].set_ylim(0, top) else: axs[2*j + 2, 0].set_ylim(0, 1) axs[2*j + 3, 0].set_ylim(0, 1) axs[2*j + 2, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[2*j + 3, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[-1, 0].set_xlabel('epochs') else: num_of_plots = len(self.metric_names) + 1 if show_metrics else 1 fig, axs = plt.subplots(num_of_plots, 1, squeeze=False, figsize=(8, 3*num_of_plots)) fig.patch.set_facecolor('w') # Loss Plots: axs[0, 0].plot(self.loss_values) axs[0, 0].set_title('Model Loss') axs[0, 0].set_ylabel('loss') axs[0, 0].set_xlabel('epochs') bottom, top = axs[0, 0].get_ylim() axs[0, 0].set_ylim(0, top) axs[0, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) # Metrics Plots: if self.metric_names and show_metrics: for j, metric_values in enumerate(self.flipped_metrics()): axs[j+1, 0].plot(metric_values) if j == 0: axs[j+1, 0].set_title('Model Metrics') axs[j+1, 0].set_ylabel(self.metric_names[j]) if len(self.metric_names[j]) == 2: bottom, top = axs[j+1, 0].get_ylim() axs[j+1, 0].set_ylim(0, top) else: axs[j+1, 0].set_ylim(0, 1) axs[j+1, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[-1, 0].set_xlabel('epochs') plt.show()Methods
def add(self, loss_value, metric_values, val_loss_value=None, val_metric_values=None)-
Add Values to the history of the training session.
Expand source code
def add(self, loss_value, metric_values, val_loss_value=None, val_metric_values=None): """Add Values to the history of the training session.""" self.loss_values.append(loss_value) self.metric_values.append(metric_values) if self.valid: self.val_loss_values.append(val_loss_value) self.val_metric_values.append(val_metric_values) def flipped_metrics(self)-
Return the metric Values as flipped for easy visualization and data extraction.
Expand source code
def flipped_metrics(self): """Return the metric Values as flipped for easy visualization and data extraction.""" return np.array(self.metric_values).T def flipped_valid_metrics(self)-
Return the metric Values as flipped for easy visualization and data extraction.
Expand source code
def flipped_valid_metrics(self): """Return the metric Values as flipped for easy visualization and data extraction.""" assert self.valid return np.concatenate([ np.array(self.metric_values).T, np.array(self.val_metric_values).T ], axis=0) def show(self, show_metrics=True)-
Show the training statistics in an image from saved data.
Expand source code
def show(self, show_metrics=True): """Show the training statistics in an image from saved data.""" if self.valid: num_of_plots = len(self.metric_names) + 1 if show_metrics else 1 num_of_plots *= 2 fig, axs = plt.subplots(num_of_plots, 1, squeeze=False, figsize=(8, 3 * num_of_plots)) fig.patch.set_facecolor('w') # Loss Plots: axs[0, 0].plot(self.loss_values) axs[0, 0].set_title('Model Loss') axs[0, 0].set_ylabel('loss') bottom, top = axs[0, 0].get_ylim() axs[0, 0].set_ylim(0, top) axs[0, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[1, 0].plot(self.val_loss_values) axs[1, 0].set_ylabel('val_loss') axs[1, 0].set_xlabel('epochs') bottom, top = axs[1, 0].get_ylim() axs[1, 0].set_ylim(0, top) axs[1, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) # Metrics Plots: if self.metric_names and show_metrics: for j, (metric_name, metric_values, valid_values) in enumerate(zip(self.metric_names, self.flipped_metrics(), self.flipped_valid_metrics())): if j == 0: axs[2*j + 2, 0].set_title('Model Metrics') axs[2*j + 2, 0].plot(metric_values) axs[2*j + 3, 0].plot(valid_values) axs[2*j + 2, 0].set_ylabel(metric_name) axs[2*j + 3, 0].set_ylabel('val_' + metric_name) if len(metric_name) == 2: bottom, top = axs[2*j + 2, 0].get_ylim() axs[2*j + 2, 0].set_ylim(0, top) bottom, top = axs[2*j + 3, 0].get_ylim() axs[2*j + 3, 0].set_ylim(0, top) else: axs[2*j + 2, 0].set_ylim(0, 1) axs[2*j + 3, 0].set_ylim(0, 1) axs[2*j + 2, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[2*j + 3, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[-1, 0].set_xlabel('epochs') else: num_of_plots = len(self.metric_names) + 1 if show_metrics else 1 fig, axs = plt.subplots(num_of_plots, 1, squeeze=False, figsize=(8, 3*num_of_plots)) fig.patch.set_facecolor('w') # Loss Plots: axs[0, 0].plot(self.loss_values) axs[0, 0].set_title('Model Loss') axs[0, 0].set_ylabel('loss') axs[0, 0].set_xlabel('epochs') bottom, top = axs[0, 0].get_ylim() axs[0, 0].set_ylim(0, top) axs[0, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) # Metrics Plots: if self.metric_names and show_metrics: for j, metric_values in enumerate(self.flipped_metrics()): axs[j+1, 0].plot(metric_values) if j == 0: axs[j+1, 0].set_title('Model Metrics') axs[j+1, 0].set_ylabel(self.metric_names[j]) if len(self.metric_names[j]) == 2: bottom, top = axs[j+1, 0].get_ylim() axs[j+1, 0].set_ylim(0, top) else: axs[j+1, 0].set_ylim(0, 1) axs[j+1, 0].xaxis.set_major_locator(MaxNLocator(integer=True)) axs[-1, 0].set_xlabel('epochs') plt.show()