jgrewe/chirp_probing
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge branch 'master' of https://whale.am28.uni-tuebingen.de/git/jgrewe/chirp_probing

Browse Source
This commit is contained in:
Jan Grewe 2020-09-24 09:08:57 +02:00
commit bdbab448ed
2 changed files with 32 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
chirps_as_probing_signals.md
View File

@ -43,11 +43,12 @@ Won't do, this is trivial?!
* Work out the difference between baseline activity and a foreign chirp response: --> done * Work out the difference between baseline activity and a foreign chirp response: --> done
* calculate the discriminability between the baseline (no-other fish present) and the another fish is present for each contrast * calculate the discriminability between the baseline (no-other fish present) and the another fish is present for each contrast
* Work out the difference between the soliloquy and the response to self generated chirp in a communication context * Work out the difference between the soliloquy and the response to self generated chirp in a communication context -> done
* Compare to the beat alone parts of the responses. * Compare to the beat alone parts of the responses. -> done
* What kernels to use? * What kernels to use?
* Duration of the chrip window? * Duration of the chrip window?
* sorting according to phase? * sorting according to phase?
* we could filter the P-unit responses to model the ELL filering
## Random thoughts ## Random thoughts
@ -55,4 +56,3 @@ Won't do, this is trivial?!
* Raab et al show this is also the case with rises. * Raab et al show this is also the case with rises.
* Check role of AFRs and rises in Tallarovic et al, Hupe et al. * Check role of AFRs and rises in Tallarovic et al, Hupe et al.
* we actually do not observe chirps without stimulation * we actually do not observe chirps without stimulation
*

44
response_discriminability.py
View File

@ -273,7 +273,7 @@ def foreign_fish_detection_beat(block_map, df, all_contrasts, all_conditions, ke
baseline_dist = within_group_distance(no_other_snippets) baseline_dist = within_group_distance(no_other_snippets)
comp_dist = across_group_distance(no_other_snippets, self_snippets) comp_dist = across_group_distance(no_other_snippets, self_snippets)
# sort and perfom roc # sort and perfom ROC analysis
triangle_indices = np.tril_indices_from(baseline_dist, -1) triangle_indices = np.tril_indices_from(baseline_dist, -1)
valid_distances_baseline = baseline_dist[triangle_indices] valid_distances_baseline = baseline_dist[triangle_indices]
temp1 = np.zeros_like(valid_distances_baseline) temp1 = np.zeros_like(valid_distances_baseline)
@ -344,7 +344,7 @@ def foreign_fish_detection_chirp(block_map, df, all_contrasts, all_conditions, k
self_vs_alone_dist = across_group_distance(alone_chirping_snippets, self_snippets) self_vs_alone_dist = across_group_distance(alone_chirping_snippets, self_snippets)
other_vs_baseline_dist = across_group_distance(baseline_snippets, other_snippets) other_vs_baseline_dist = across_group_distance(baseline_snippets, other_snippets)
# sort and perfom roc for two comparisons # sort and perfom ROC analysis for two comparisons
# 1. soliloquy vs. self chirping in company # 1. soliloquy vs. self chirping in company
# 2. other chirping vs. nobody is chirping # 2. other chirping vs. nobody is chirping
triangle_indices = np.tril_indices_from(alone_chirping_dist, -1) triangle_indices = np.tril_indices_from(alone_chirping_dist, -1)
@ -381,7 +381,7 @@ def foreign_fish_detection_chirp(block_map, df, all_contrasts, all_conditions, k
return detection_performances return detection_performances
def plot_detection_results(data_frame, df, kernel_width, cell): def plot_detection_results(data_frame, df, kernel_width, cell, figure_name=None):
cell_results = data_frame[(data_frame.cell == cell) & (data_frame.df == df)] cell_results = data_frame[(data_frame.cell == cell) & (data_frame.df == df)]
conditions = sorted(cell_results.detection_task.unique()) conditions = sorted(cell_results.detection_task.unique())
kernels = sorted(cell_results.kernel_width.unique()) kernels = sorted(cell_results.kernel_width.unique())
@ -390,26 +390,27 @@ def plot_detection_results(data_frame, df, kernel_width, cell):
fig_grid = (8, 7) fig_grid = (8, 7)
for i, c in enumerate(conditions): for i, c in enumerate(conditions):
condition_results = cell_results[cell_results.detection_task == c] condition_results = cell_results[cell_results.detection_task == c]
roc_data = condition_results[condition_results.kernel_width == kernel_width]
contrasts = roc_data.contrast.unique()
roc_ax = plt.subplot2grid(fig_grid, (i * 2 + i, 0), colspan=3, rowspan=2) roc_ax = plt.subplot2grid(fig_grid, (i * 2 + i, 0), colspan=3, rowspan=2)
roc_ax.set_title(c, fontsize=9, ha="left")
auc_ax = plt.subplot2grid(fig_grid, (i * 2 + i, 4), colspan=3, rowspan=2) auc_ax = plt.subplot2grid(fig_grid, (i * 2 + i, 4), colspan=3, rowspan=2)
roc_data = condition_results[condition_results.kernel_width == kernel_width]
contrasts = roc_data.contrast.unique()
for c in contrasts: for c in contrasts:
tpr = roc_data.true_positives[roc_data.contrast == c].values[0] tpr = roc_data.true_positives[roc_data.contrast == c].values[0]
fpr = roc_data.false_positives[roc_data.contrast == c].values[0] fpr = roc_data.false_positives[roc_data.contrast == c].values[0]
roc_ax.plot(fpr, tpr, label="%.3f" % c, zorder=2) roc_ax.plot(fpr, tpr, label="%.3f" % c, zorder=2)
roc_ax.legend(loc="best", fontsize=6, ncol=2, frameon=False) roc_ax.legend(loc="best", fontsize=6, ncol=2, frameon=False)
roc_ax.plot([0., 1.],[0., 1.], color="k", lw=0.5, ls="--", zorder=0) roc_ax.plot([0., 1.],[0., 1.], color="k", lw=0.5, ls="--", zorder=0)
roc_ax.set_xlabel("false positive rate", fontsize=9)
roc_ax.set_ylabel("true positive rate", fontsize=9)
roc_ax.set_xticks(np.arange(0.0, 1.01, 0.5)) roc_ax.set_xticks(np.arange(0.0, 1.01, 0.5))
roc_ax.set_xticks(np.arange(0.0, 1.01, 0.25), minor=True) roc_ax.set_xticks(np.arange(0.0, 1.01, 0.25), minor=True)
roc_ax.set_xticklabels(np.arange(0.0, 1.01, 0.5), fontsize=8)
roc_ax.set_yticks(np.arange(0.0, 1.01, 0.5)) roc_ax.set_yticks(np.arange(0.0, 1.01, 0.5))
roc_ax.set_yticks(np.arange(0.0, 1.01, 0.25), minor=True) roc_ax.set_yticks(np.arange(0.0, 1.01, 0.25), minor=True)
if i == len(conditions) - 1:
roc_ax.set_xticklabels(np.arange(0.0, 1.01, 0.5), fontsize=8)
roc_ax.set_xlabel("false positive rate", fontsize=9)
roc_ax.set_ylabel("true positive rate", fontsize=9)
roc_ax.set_yticklabels(np.arange(0.0, 1.01, 0.5), fontsize=8) roc_ax.set_yticklabels(np.arange(0.0, 1.01, 0.5), fontsize=8)
for k in kernels: for k in kernels:
@ -417,19 +418,25 @@ def plot_detection_results(data_frame, df, kernel_width, cell):
aucs = np.asarray(condition_results.auc[condition_results.kernel_width == k]) aucs = np.asarray(condition_results.auc[condition_results.kernel_width == k])
aucs_sorted = aucs[np.argsort(contrasts)] aucs_sorted = aucs[np.argsort(contrasts)]
contrasts_sorted = np.sort(contrasts) contrasts_sorted = np.sort(contrasts)
auc_ax.plot(contrasts_sorted, aucs_sorted, marker=".", label=r"$\sigma$: %.4f" % k) auc_ax.plot(contrasts_sorted, aucs_sorted, marker=".", label=r"$\sigma$: %.4f" % k, zorder=1)
if i == len(conditions) - 1:
auc_ax.set_xlabel("contrast [%]") auc_ax.set_xlabel("contrast [%]", fontsize=9)
else:
auc_ax.set_xticklabels("")
auc_ax.set_ylim([0.25, 1.0]) auc_ax.set_ylim([0.25, 1.0])
auc_ax.set_ylabel("discriminability") auc_ax.set_yticks(np.arange(0.25, 1.01, 0.25))
auc_ax.set_yticklabels(np.arange(0.25, 1.01, 0.25), fontsize=8)
auc_ax.set_ylabel("discriminability", fontsize=9)
auc_ax.legend(ncol=2, fontsize=6, handletextpad=0.4, columnspacing=1.0, labelspacing=0.25) auc_ax.legend(ncol=2, fontsize=6, handletextpad=0.4, columnspacing=1.0, labelspacing=0.25)
auc_ax.plot([min(contrasts), max(contrasts)], [0.5, 0.5], lw=0.5, ls"--",) auc_ax.plot([min(contrasts), max(contrasts)], [0.5, 0.5], lw=0.5, ls="--", zorder=0)
fig.savefig("discrimination.pdf") name = figure_name if figure_name is not None else "foreign_fish_detection.pdf"
name = (name + ".pdf") if ".pdf" not in name else name
plt.savefig(os.path.join(figure_folder, name))
def foreign_fish_detection(block_map, all_dfs, all_contrasts, all_conditions, current_df=None, cell_name=""): def foreign_fish_detection(block_map, all_dfs, all_contrasts, all_conditions, current_df=None, cell_name=""):
dfs = [current_df] if current_df is not None else all_dfs dfs = [current_df] if current_df is not None else all_dfs
kernels = [0.00025, 0.0005, 0.001, 0.0025, 0.005] kernels = [0.00025, 0.0005, 0.001, 0.0025]
result_dicts = [] result_dicts = []
for df in dfs: for df in dfs:
for kw in kernels: for kw in kernels:
@ -447,6 +454,11 @@ def foreign_fish_detection(block_map, all_dfs, all_contrasts, all_conditions, cu
return result_dicts return result_dicts
def estimate_chirp_phase(am, chirp_times):
pass
def process_cell(filename, dfs=[], contrasts=[], conditions=[]): def process_cell(filename, dfs=[], contrasts=[], conditions=[]):
nf = nix.File.open(filename, nix.FileMode.ReadOnly) nf = nix.File.open(filename, nix.FileMode.ReadOnly)
block_map, all_contrasts, all_dfs, all_conditions = sort_blocks(nf) block_map, all_contrasts, all_dfs, all_conditions = sort_blocks(nf)