[Fix] Fix wrong condition judgment in analyze_logs.py and prevent empty curve.

[imyhxy]

Motivation

Make analyze_log.py works with single validation iteration. Currently, the analyza_logs.py use iters = iters[:-1] to get rid of the last validation iteration, but in a very small val set, there maybe only one validation step. In such case, the analyze_log.py won't work, it use print out a blank figure. This PR fixed that.

Modification

When there is only one item in the iters list, then don't slice it.

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects, like MMDet or MMSeg.
• CLA has been signed and all committers have signed the CLA in this PR.

[imyhxy]

Test with this log file.

python tools/analysis_tools/analyze_logs.py plot_curve 20211028_115735.log.json --keys accuracy_top-1

20211028_115735.log

[codecov]

Codecov Report

Merging #510 (1bacb05) into master (72cffac) will increase coverage by 0.05%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #510      +/-   ##
==========================================
+ Coverage   79.42%   79.48%   +0.05%     
==========================================
  Files         106      106              
  Lines        5959     5975      +16     
  Branches      962      968       +6     
==========================================
+ Hits         4733     4749      +16     
  Misses       1095     1095              
  Partials      131      131              

Flag	Coverage Δ
unittests	79.48% <ø> (+0.05%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmcls/utils/logger.py	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 72cffac...1bacb05. Read the comment docs.

[mzr1996]

Test with this log file.

python tools/analysis_tools/analyze_logs.py plot_curve 20211028_115735.log.json --keys accuracy_top-1

20211028_115735.log

Thanks for your contribution.
Here, I think you want to put the log json file, but put a log file by mistake.

[imyhxy]

20211028_115735.log.json.txt
@mzr1996 Please remove the .txt suffix because github not supported raw json.

[mzr1996]

I have read your modification.
First, let me confirm your need. You want to analyze the log and plot the accuracy curve, right?
Let me show the code:

            if 'mAP' in metric:    # <-- Here, it's the condition to find the metric in val steps
                xs = np.arange(1, max(epochs) + 1)
                ys = []
                for epoch in epochs:
                    ys += log_dict[epoch][metric]
                ax = plt.gca()
                ax.set_xticks(xs)
                plt.xlabel('epoch')
                plt.plot(xs, ys, label=legend[i * num_metrics + j], marker='o')
            else:                  # <-- Else, it will find the metric in train steps
                xs = []
                ys = []
                num_iters_per_epoch = log_dict[epochs[0]]['iter'][-1]
                for epoch in epochs:
                    iters = log_dict[epoch]['iter']
                    if log_dict[epoch]['mode'][-1] == 'val':
                        iters = iters[:-1]
                    xs.append(
                        np.array(iters) + (epoch - 1) * num_iters_per_epoch)
                    ys.append(np.array(log_dict[epoch][metric][:len(iters)]))
                xs = np.concatenate(xs)
                ys = np.concatenate(ys)
                plt.xlabel('iter')
                plt.plot(
                    xs, ys, label=legend[i * num_metrics + j], linewidth=0.5)

The problem is not the last eval step is skipped, but we forget to add "accuracy" to the condition to find the metric in val steps.
So I think a better solution is to modify the first line to:

            if 'mAP' in metric or 'accuracy' in metric:

And the original code can only work when evaluate in every epoch. So please change the original code to:

            if 'mAP' in metric or 'accuracy' in metric:
                xs = []
                ys = []
                for epoch in epochs:
                    xs.append(epoch)
                    ys.extend(log_dict[epoch][metric])
                ax = plt.gca()
                ax.set_xticks(xs)
                plt.xlabel('epoch')
                plt.plot(xs, ys, label=legend[i * num_metrics + j], marker='o')

[imyhxy]

Yeah, you are right. I have modified the code according to your suggestion. I think the iters thing still need a sentinel value to prevent it fall into an empty list.

[mzr1996]

LGTM, thanks for your contribution 😃