confusion mat and predictions

confusion_and_prediction.py

from pybmi.modeling.loading.benchmarks import EventsBenchmarkLoader
from cogdata_service.service import simple_api
from workflow.experiments.utils import partition
from pybmi.modeling.events.metrics import events_confusion
import json

import matplotlib.pyplot as plt

def plot_preds(preds, time, alpha=1, ax=None, c='k', events=None):
    if ax is None:
        fig, ax = plt.subplots(1, 1, figsize=[15, 4])
        fig.patch.set_facecolor('white')
        
    offsets = 1.1*np.arange(preds.shape[-1])
    pr = preds- offsets[np.newaxis]
    ax.plot(time, pr, c=c, alpha=alpha, lw=1)
    ax.spines['top'].set_visible(False)
    ax.spines['right'].set_visible(False)
    ax.spines['left'].set_visible(False)
    ax.set_yticks(-offsets + 0.5)
    if events is not None:
        ax.set_yticklabels(events)
    ax.set_xlim(time[0], time[-1])
    ax.set_xlabel('Time (s)')
    return ax


api = simple_api.API()

model_id = '61e19360-942d-47c5-8c79-f761b56cde1d'

loader = EventsBenchmarkLoader(config={'model_predictions': {'model_id': model_id}})

m = api.get_model(model_id)
config = m['config']
datas = partition(loader, config['data']['validation'][0])

preds = np.concatenate([data.main['model predictions'] for data in datas], axis=0)
labels = np.concatenate([data.main['events'] for data in datas], axis=0)
time = np.concatenate([data.main['time'] for data in datas], axis=0)

scores = events_confusion(preds, labels, threshold=0.4)

ev_names = json.loads(datas[0].metadata['events_names'])

fig = plt.figure(figsize=[4, 4])
fig.patch.set_facecolor('white')
sns.heatmap(scores, annot=True, fmt='.0f',
            xticklabels=ev_names + ['null'], yticklabels=ev_names + ['null'],
            cbar=False, square=True, cmap='Greys')
plt.ylabel('True Event')
plt.xlabel('Predicted Event');


ax = plot_preds(preds, time, events=ev_names, c='r')
plot_preds(labels, time, events=ev_names, c='k', ax=ax, alpha=0.5)