metric-visualizer 0.4.32

No project description provided

MetricVisualizer - for easy managing performance metric

Automated metric visualization for comparison experiments

• Box plot
• Trajectory plot
• Scatter plot
• Bar plot
• Violin plot
• Rank test
• On the way

Install

If you want to make tikz(latex) plots, you need to install texlive (other latex release version are not tested).

pip install metric_visualizer

用法说明 Usage

假设存在多组对比实验(或者一组参数设置)，则称之为trial，每组实验存在多个metric(例如AUC，Accuracy，F1，Loss等)， 每组参照实验重复n词，则使用以下方法监听实验结果： Assume that there exist multiple sets of comparison experiments (or a set of parameter settings), called trials, with multiple metrics (e.g., AUC, accuracy, F1, loss, etc.) for each set of experiments. Repeat n words for each set of reference experiments, and then listen to the results of the experiments using the following method.

import numpy as np
from metric_visualizer import MetricVisualizer

MV = MetricVisualizer()

trial_num = 5  # number of different trials,
repeat = 10  # number of repeats
metric_num = 3  # number of metrics

for trial in range(trial_num):
    for r in range(repeat):  # repeat the experiments to plot violin or box figure
        metrics = [(np.random.random() + n) for n in range(metric_num)]  # n is metric scale factor
        for i, m in enumerate(metrics):
            MV.add_metric('Metric-{}'.format(i + 1), round(m, 2))  # Add metric by metric name
    MV.next_trial()  # move to next trial

画图代码如下：

save_prefix = None
MV.summary(save_path=save_prefix, no_print=True)  # save fig into .tex and .pdf format
MV.traj_plot_by_trial(save_name=save_prefix, xlabel='', xrotation=30, minorticks_on=True)  # save fig into .tex and .pdf format
MV.violin_plot_by_trial(save_name=save_prefix)  # save fig into .tex and .pdf format
MV.box_plot_by_trial(save_name=save_prefix)  # save fig into .tex and .pdf format
MV.avg_bar_plot_by_trial(save_name=save_prefix)  # save fig into .tex and .pdf format
MV.sum_bar_plot_by_trial(save_name=save_prefix)  # save fig into .tex and .pdf format

# 此函数适合对比不同模型性能，每个模型代表一个trial，综合多个metric进行Scott-Knott Rank Test，并绘制箱型图
MV.scott_knott_plot(save_name=save_prefix, minorticks_on=False)  

print(MV.rank_test_by_trail('trial0'))  # save fig into .tex and .pdf format
print(MV.rank_test_by_metric('metric1'))  # save fig into .tex and .pdf format


# save_path = None
# MV.summary(save_path=save_path)  # save fig into .tex and .pdf format
# MV.traj_plot_by_metric(save_path=save_path, xlabel='', xrotation=30)  # save fig into .tex and .pdf format
# MV.violin_plot_by_metric(save_path=save_path)  # save fig into .tex and .pdf format
# MV.box_plot_by_metric(save_path=save_path)  # save fig into .tex and .pdf format
# MV.avg_bar_plot_by_metric(save_path=save_path)  # save fig into .tex and .pdf format
# MV.sum_bar_plot_by_metric(save_path=save_path)  # save fig into .tex and .pdf format

 -------------------- Metric Summary --------------------
╒══════════╤═════════╤══════════════════════════════════════════════════════════════╤═════════════════════════════════════════════════════════════╕
│ Metric   │ Trial   │ Values                                                       │ Summary                                                     │
╞══════════╪═════════╪══════════════════════════════════════════════════════════════╪═════════════════════════════════════════════════════════════╡
│ Metric-1 │ trial-0 │ [0.35, 0.65, 0.67, 0.51, 0.04, 0.43, 0.46, 0.58, 0.11, 0.66] │ ['Avg:0.45, Median: 0.48, IQR: 0.22, Max: 0.67, Min: 0.04'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-1 │ trial-1 │ [0.52, 0.1, 0.11, 0.86, 0.49, 0.7, 0.77, 0.96, 0.16, 0.65]   │ ['Avg:0.53, Median: 0.58, IQR: 0.41, Max: 0.96, Min: 0.1']  │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-1 │ trial-2 │ [0.73, 0.99, 0.13, 0.72, 0.63, 0.61, 0.14, 0.85, 0.71, 0.86] │ ['Avg:0.64, Median: 0.72, IQR: 0.17, Max: 0.99, Min: 0.13'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-1 │ trial-3 │ [0.99, 0.69, 0.86, 0.2, 0.4, 0.1, 0.05, 0.07, 0.95, 0.31]    │ ['Avg:0.46, Median: 0.36, IQR: 0.62, Max: 0.99, Min: 0.05'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-1 │ trial-4 │ [0.58, 0.95, 0.73, 0.63, 0.04, 0.19, 0.5, 0.9, 0.64, 0.89]   │ ['Avg:0.6, Median: 0.64, IQR: 0.27, Max: 0.95, Min: 0.04']  │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-2 │ trial-0 │ [1.58, 1.32, 1.98, 1.76, 1.31, 1.6, 1.6, 1.22, 1.3, 1.19]    │ ['Avg:1.49, Median: 1.45, IQR: 0.29, Max: 1.98, Min: 1.19'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-2 │ trial-1 │ [1.88, 1.67, 1.77, 1.94, 1.01, 1.6, 1.25, 1.63, 1.62, 1.91]  │ ['Avg:1.63, Median: 1.65, IQR: 0.21, Max: 1.94, Min: 1.01'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-2 │ trial-2 │ [1.4, 1.94, 1.28, 1.78, 1.01, 1.08, 1.21, 1.82, 1.78, 1.18]  │ ['Avg:1.45, Median: 1.34, IQR: 0.59, Max: 1.94, Min: 1.01'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-2 │ trial-3 │ [1.79, 1.35, 1.14, 1.5, 1.73, 1.06, 1.98, 1.75, 1.07, 1.49]  │ ['Avg:1.49, Median: 1.5, IQR: 0.49, Max: 1.98, Min: 1.06']  │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-2 │ trial-4 │ [1.93, 1.81, 1.18, 1.08, 1.57, 1.85, 1.95, 1.94, 1.58, 1.35] │ ['Avg:1.62, Median: 1.7, IQR: 0.43, Max: 1.95, Min: 1.08']  │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-3 │ trial-0 │ [2.85, 2.87, 2.3, 2.05, 2.86, 2.34, 2.85, 2.3, 2.95, 2.53]   │ ['Avg:2.59, Median: 2.69, IQR: 0.54, Max: 2.95, Min: 2.05'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-3 │ trial-1 │ [2.31, 2.41, 2.34, 2.96, 2.48, 2.68, 2.99, 2.94, 2.01, 2.46] │ ['Avg:2.56, Median: 2.47, IQR: 0.44, Max: 2.99, Min: 2.01'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-3 │ trial-2 │ [2.65, 2.5, 2.68, 2.34, 2.32, 2.61, 2.61, 2.88, 2.86, 2.36]  │ ['Avg:2.58, Median: 2.61, IQR: 0.24, Max: 2.88, Min: 2.32'] │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-3 │ trial-3 │ [2.29, 2.12, 2.4, 2.81, 2.5, 2.54, 2.82, 2.61, 2.45, 2.44]   │ ['Avg:2.5, Median: 2.48, IQR: 0.16, Max: 2.82, Min: 2.12']  │
├──────────┼─────────┼──────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ Metric-3 │ trial-4 │ [2.41, 2.12, 2.31, 2.29, 2.46, 2.95, 2.74, 2.66, 2.34, 2.65] │ ['Avg:2.49, Median: 2.44, IQR: 0.33, Max: 2.95, Min: 2.12'] │
╘══════════╧═════════╧══════════════════════════════════════════════════════════════╧═════════════════════════════════════════════════════════════╛
 -------------------- Metric Summary --------------------

Plot via Matplotlib (or Tikz)

Traj Plot tikz version

Box Plot tikz version

Violin Plot tikz version

Average Bar Plot tikz version

Sum Bar Plot tikz version

Real Usage Example in PyABSA

To analyze the impact of max_seq_len, we can use MetricVisualizer as following:

pip install pyabsa  # install pyabsa

import random
import os
from metric_visualizer import MetricVisualizer

from pyabsa.functional import Trainer
from pyabsa.functional import APCConfigManager
from pyabsa.functional import ABSADatasetList
from pyabsa.functional import APCModelList

config = APCConfigManager.get_config()
config.model = APCModelList.FAST_LCF_BERT
config.lcf = 'cdw'
config.seed = [random.randint(0, 10000) for _ in range(3)] # each trial repeats with different seed

MV = MetricVisualizer()
config.MV = MV

max_seq_lens = [60, 70, 80, 90, 100]

for max_seq_len in max_seq_lens:
    config.max_seq_len = max_seq_len
    dataset = ABSADatasetList.Laptop14
    Trainer(config=config,
            dataset=dataset,  # train set and test set will be automatically detected
            auto_device=True  # automatic choose CUDA or CPU
            )
    config.MV.next_trial()

save_prefix = os.getcwd()
MV.summary(save_path=save_prefix, no_print=True)  # save fig into .tex and .pdf format

 # save fig into .tex and .pdf format
MV.traj_plot_by_trial(save_path=save_prefix, xticks=max_seq_lens) 
MV.violin_plot_by_trial(save_path=save_prefix, xticks=max_seq_lens) 
MV.box_plot_by_trial(save_path=save_prefix, xticks=max_seq_lens) 
MV.avg_bar_plot_by_trial(save_path=save_prefix, xticks=max_seq_lens) 
MV.sum_bar_plot_by_trial(save_path=save_prefix, xticks=max_seq_lens) 
MV.scott_knott_plot(save_path=save_prefix, xticks=max_seq_lens, minorticks_on=False)