Reorganize folders

MEArec_SqMEA1015/README.md

@@ -0,0 +1,14 @@
+# Comparison of spike sorters with MEArec and SpikeInterface
+
+@author : Samuel Garcia
+
+
+Dataset generated with [MEArec](https://github.com/alejoe91/MEArec) (author: Alessio Paolo Buccino)
+
+The recordings are simulated on a **SqMEA-10-15** probe, with 100 channels in a 10x10 configurtion, and with an inter-electrode distance of 15 um. There are 50 ground-truth neurons and the duration is 60 seconds.
+
+The file **recordings_50cells_SqMEA-10-15um_60.0_10.0uV_27-03-2019_13_31.h5** is freely downloadable [here](https://doi.org/10.5281/zenodo.3260283). You can download the file, unzip it in this folder, and run the notebook.
+
+
+This notebook is a demonstration of the SpikeInterface ground-truth comparison framework.
+The notebook launches several popular sorters on the recording and computes several performence metrics.

MEArec_SqMEA1015/main_script.py

@@ -0,0 +1,124 @@
+import os
+
+import pandas as pd
+import matplotlib.pyplot as plt
+import seaborn as sns
+import spikeextractors as se
+from spiketoolkit.study import GroundTruthStudy
+import spikewidgets as sw
+from pathlib import Path
+
+plot_mearec = False
+
+def clean_plot(ax, label, sorters):
+    ax.set_xlabel('')
+    ax.set_ylabel(label, fontsize=20)
+    ax.yaxis.set_tick_params(labelsize=12)
+    ax.set_xticklabels(sorters, fontsize=15, rotation=45, ha='center')
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+    fig = ax.get_figure()
+    fig.subplots_adjust(bottom=0.2)
+    fig.set_size_inches(8, 7)
+
+p = Path('.')
+
+study_folder = p / 'study_mearec_SqMEA1015um'
+
+mearec_filename = p / 'recordings_50cells_SqMEA-10-15um_60.0_10.0uV_27-03-2019_13-31.h5'
+rec0 = se.MEArecRecordingExtractor(mearec_filename)
+gt_sorting0 = se.MEArecSortingExtractor(mearec_filename)
+study = None
+
+if plot_mearec:
+    import MEArec as mr
+    import MEAutility as mu
+
+    recgen = mr.load_recordings(mearec_filename)
+
+    # plot the probe
+    probe_info = recgen.info['electrodes']
+    probe_name = recgen.info['electrodes']['electrode_name']
+    mea = mu.return_mea(info=probe_info)
+    fig1, ax1 = plt.subplots()
+    mu.plot_probe(mea, ax=ax1)
+    ax1.axis('off')
+
+    fig2, axs = plt.subplots(nrows=3, ncols=4)
+    for i, unit_id in enumerate(gt_sorting0.get_unit_ids()[:12]):
+        ax = axs.flatten()[i]
+        mr.plot_templates(recgen, template_ids=unit_id, ax=ax)
+        ax.set_title('unit #{}'.format(unit_id))
+
+    fig3, ax3 = plt.subplots()
+    mr.plot_recordings(recgen, start_time=0, end_time=5, lw=0.1, ax=ax3)
+
+# Setup study folder
+if not study_folder.is_dir():
+    print('Setting up study folder:', study_folder)
+    gt_dict = {'rec0': (rec0, gt_sorting0)}
+    study = GroundTruthStudy.create(study_folder, gt_dict)
+
+if study is None:
+    study = GroundTruthStudy(study_folder)
+
+# Run sorters
+sorter_list = ['herdingspikes', 'kilosort2', 'ironclust',
+               'spykingcircus', 'tridesclous']
+sorter_params = {'mountainsort4': {'adjacency_radius': 50},
+                 'spykingcircus': {'adjacency_radius': 50}}
+
+if not (study_folder / 'sorter_folders').is_dir():
+    print('Running sorters')
+    study.run_sorters(sorter_list, mode='keep', engine='loop')
+
+# Perform comparisons
+print('Performing comparisons and getting results')
+study.run_comparisons(exhaustive_gt=True, compute_misclassification=True)
+
+comparisons = study.comparisons
+dataframes = study.aggregate_dataframes(accuracy=0.8)
+
+# Plot confusion matrices
+fig4 = plt.figure()
+for i, ((rec_name, sorter_name), comp) in enumerate(comparisons.items()):
+    print(rec_name, sorter_name)
+    comp.print_summary(accuracy=0.8)
+
+    ax = fig4.add_subplot(2, 3, i + 1)
+    sw.plot_confusion_matrix(comp, ax=ax, count_text=False, unit_ticks=False)
+    ax.set_title(sorter_name)
+
+plt.show()
+
+run_times = dataframes['run_times']
+perf_units = dataframes['perf_by_units']
+perf_avg = dataframes['perf_pooled_with_average']
+count_units = dataframes['count_units']
+
+exclude_sorters = ['mountainsort4',  'tridesclous']
+
+if len(exclude_sorters) > 0:
+    for name in exclude_sorters:
+        run_times = run_times[run_times['sorter_name'] != name]
+        perf_units = perf_units[perf_units['sorter_name'] != name]
+        perf_avg = perf_avg[perf_avg['sorter_name'] != name]
+        count_units = count_units[count_units['sorter_name'] != name]
+
+sorters = run_times['sorter_name']
+
+fig5, ax5 = plt.subplots()
+ax5 = sns.barplot(data=dataframes['run_times'], x='sorter_name', y='run_time', ax=ax5, order=sorters)
+fig6, ax6 = plt.subplots()
+ax6 = sns.swarmplot(data=dataframes['perf_by_units'], x='sorter_name', y='accuracy', ax=ax6, order=sorters)
+fig7, ax7 = plt.subplots()
+ax7 = sns.barplot(data=dataframes['perf_pooled_with_average'], x='sorter_name', y='precision', ax=ax7, order=sorters)
+fig8, ax8 = plt.subplots()
+ax8 = sns.barplot(data=dataframes['count_units'], x='sorter_name', y='num_bad', ax=ax8, order=sorters)
+
+sorters = run_times['sorter_name']
+
+clean_plot(ax5, 'run times (s)', sorters)
+clean_plot(ax6, 'accuracy', sorters)
+clean_plot(ax7, 'precision', sorters)
+clean_plot(ax8, 'bad units', sorters)