import os
import numpy as np
import pandas as pd
[docs]class SubclassMetricsCache(object):
"""Used to calc optimized parameteres for different types of cells using existing data that contains cell responses
to different types of stimulus. The existing data should be stored in a single directory with a csv file for reach
cell type.
usage:
subcell_metrics = SubclassMetricsCache('/path/to/observations/')
"""
def __init__(self, saved_metrics_dir=None):
self._saved_metrics_dir = saved_metrics_dir if saved_metrics_dir is not None \
else os.path.join(os.path.dirname(__file__), 'cell_metrics')
self._metrics_cache = {} # subclass_name --> metrics Dataframe, stores calculated params for each cell class
@property
def saved_metrics_dir(self):
return self._saved_metrics_dir
@saved_metrics_dir.setter
def saved_metrics_dir(self, path):
if not os.path.isdir(path):
raise FileNotFoundError('{} directory not found.'.format(path))
self._saved_metrics_dir = path
[docs] def get_metrics(self, subclass_type):
"""Calculates various parameters for a given cell class, using observations stored in saved_metrics_dir
:param subclass_type: name of cell subclass types
:return: A dictionary of parameters for various types of inputs
"""
if subclass_type in self._metrics_cache:
# See if parameters for cell type have already been calculated
return self._metrics_cache[subclass_type]
else:
subtype_metrics = self._load_metrics(subclass_type)
self._metrics_cache[subclass_type] = subtype_metrics
return subtype_metrics
def _load_metrics(self, ctype):
# Helpter function for get_metrics, calculates subtype parameters for the given cell-class type (ctype)
if ctype.find('_sus') >= 0:
prs_fn = os.path.join(self.saved_metrics_dir, '{}_cells_v3.csv'.format(ctype))
else:
prs_fn = os.path.join(self.saved_metrics_dir, '{}_cell_data.csv'.format(ctype))
prs_df = pd.read_csv(prs_fn)
N_class, nmet = np.shape(prs_df)
# Group data by subclasses based on max F0 vals
exp_df = prs_df.iloc[:, [13, 14, 17, 18, 28, 45, 46, 47, 48, 49, 50, 51, 52, 53,
54]].copy() # Bl_lat,Wh_lat,Bl_si, wh_si, spont, 5 F0s, 5 F1s
sub_df = exp_df.iloc[:, [5, 6, 7, 8, 9]]
exp_df['max_tf'] = sub_df.idxmax(axis=1).values # sub_df.idxmax(axis=1)
exp_means = exp_df.groupby(['max_tf']).mean()
exp_std = exp_df.groupby(['max_tf']).std()
exp_nsub = exp_df.groupby(['max_tf']).size()
max_ind_arr = np.where(exp_nsub == np.max(exp_nsub))
max_nsub_ind = max_ind_arr[0][0]
# Get means and std dev for subclasses
exp_prs_dict = {}
for scn in np.arange(len(exp_nsub)):
f0_exp = exp_means.iloc[scn, 5:10].values
f1_exp = exp_means.iloc[scn, 10:].values
spont_exp = exp_means.iloc[scn, 4:5].values
if ctype.find('OFF') >= 0:
si_exp = exp_means.iloc[scn, 2:3].values
ttp_exp = exp_means.iloc[scn, 0:1].values
elif ctype.find('ON') >= 0:
si_exp = exp_means.iloc[scn, 3:4].values
ttp_exp = exp_means.iloc[scn, 1:2].values
else:
si_exp = np.NaN * np.ones((1, 5))
ttp_exp = np.NaN * np.ones((1, 2))
nsub = exp_nsub.iloc[scn]
if nsub == 1:
f0_std = np.mean(exp_std.iloc[max_nsub_ind, 5:10].values) * np.ones((1, 5))
f1_std = np.mean(exp_std.iloc[max_nsub_ind, 10:].values) * np.ones((1, 5))
spont_std = np.mean(exp_std.iloc[max_nsub_ind, 4:5].values) * np.ones((1, 5))
if ctype.find('OFF') >= 0:
si_std = np.mean(exp_std.iloc[max_nsub_ind, 2:3].values) * np.ones((1, 5))
elif ctype.find('ON') >= 0:
si_std = np.mean(exp_std.iloc[max_nsub_ind, 3:4].values) * np.ones((1, 5))
else:
si_std = np.NaN * np.ones((1, 5))
else:
f0_std = exp_std.iloc[scn, 5:10].values
f1_std = exp_std.iloc[scn, 10:].values
spont_std = exp_std.iloc[scn, 4:5].values
if ctype.find('OFF') >= 0:
si_std = exp_std.iloc[scn, 2:3].values
elif ctype.find('ON') >= 0:
si_std = exp_std.iloc[scn, 3:4].values
else:
si_std = np.NaN * np.ones((1, 5))
if ctype.find('t') >= 0:
tcross = 40.
si_inf_exp = (si_exp - tcross / 200.) * (200. / (200. - tcross - 40.))
elif ctype.find('s') >= 0:
tcross = 60.
si_inf_exp = (si_exp - tcross / 200.) * (200. / (200. - tcross - 40.))
else:
si_inf_exp = np.nan
dict_key = exp_means.iloc[scn].name[3:]
exp_prs_dict[dict_key] = {}
exp_prs_dict[dict_key]['f0_exp'] = f0_exp
exp_prs_dict[dict_key]['f1_exp'] = f1_exp
exp_prs_dict[dict_key]['spont_exp'] = spont_exp
exp_prs_dict[dict_key]['si_exp'] = si_exp
exp_prs_dict[dict_key]['si_inf_exp'] = si_inf_exp
exp_prs_dict[dict_key]['ttp_exp'] = ttp_exp
exp_prs_dict[dict_key]['f0_std'] = f0_std
exp_prs_dict[dict_key]['f1_std'] = f1_std
exp_prs_dict[dict_key]['spont_std'] = spont_std
exp_prs_dict[dict_key]['si_std'] = si_std
exp_prs_dict[dict_key]['nsub'] = nsub
exp_prs_dict[dict_key]['N_class'] = N_class
return exp_prs_dict
# Creates/retrives a singleton for every difference
_instances = {}
[docs] @classmethod
def load_dir(cls, saved_metrics_dir=None):
if saved_metrics_dir in cls._instances:
return cls._instances[saved_metrics_dir]
else:
metrics = cls(saved_metrics_dir)
cls._instances[saved_metrics_dir] = metrics
return metrics
[docs]def get_data_metrics_for_each_subclass(ctype):
metrics = SubclassMetricsCache.load_dir()
return metrics.get_metrics(ctype)
