def compute_all_diffs_dfs(args: Any, silent: bool = False) -> Path:
_, diffs_out = stats_fnames(args, args.normalize, extension="csv")
all_diffs_out = DATA_ROOT / diffs_out
DUDS = [
"control_pre_v_parkinsons", "park_pre_v_parkinsons",
"control_v_park_pre", "control_v_control_pre"
]
all_diffs = []
datasets = DATASETS_FULLPRE if args.fullpre else DATASETS
for dataset_name, dataset in datasets.items():
if dataset_name == "SINGLE_SUBJECT":
continue
precompute_dataset(dataset_name,
args=args,
force_all=False,
silent=silent)
pairings = Pairings(args, dataset_name)
for pair in pairings.pairs:
if pair.label in DUDS:
continue
diffs = pair.paired_differences(args)
all_diffs.append(diffs)
diffs_df = pd.concat(all_diffs)
diffs_df.to_csv(all_diffs_out)
if not silent:
print(diffs_df)
print(f"Saved all differences to {all_diffs_out}")
return all_diffs_out
def plot_raw_eigs(args: Any, show: bool = False) -> None:
"""Create violin plots of the distributions of the largest eigenvalues
Parameters
----------
args: Args
Contains the unfolding, trimming, normalization, etc options defined in
run.py
show: bool
If False (default) just save the plot. Otherwise, call plt.show() and do
NOT save
"""
if not show:
outdir = DATA_ROOT / "plots"
os.makedirs(outdir, exist_ok=True)
datasets = DATASETS_FULLPRE if args.fullpre else DATASETS
for dataset_name, subgroups in datasets.items():
for subgroup_name, eigpaths in subgroups.items():
pass
# for subgroupname, eigpaths in dataset.items():
for dataset_name in datasets:
dfs = []
for groupname, observables in precompute_dataset(dataset_name, args, silent=True).items():
path = observables["largest"]
df_full = pd.read_pickle(path)
df = pd.DataFrame(df_full.loc["largest", :], dtype=float)
n = len(df)
df["subgroup"] = [groupname for _ in range(n)]
dfs.append(df)
df = pd.concat(dfs)
sbn.set_context("paper")
sbn.set_style("ticks")
args_prefix = argstrings_from_args(args)[0]
dname = dataset_name.upper()
prefix = f"{dname}_{args_prefix}_{'fullpre_' if args.fullpre else ''}"
title = f"{dname} - Largest λ"
subtitle = args_prefix.replace("_", " ")
with sbn.axes_style("ticks"):
subfontsize = 12
fontsize = 16
ax = sbn.violinplot(x="subgroup", y="largest", data=df)
sbn.despine(offset=10, trim=True)
plt.gcf().suptitle(title, fontsize=fontsize)
ax.set_title(subtitle, fontdict={"fontsize": subfontsize})
ax.set_title(title)
ax.set_xlabel("λ_max", fontdict={"fontsize": fontsize})
ax.set_ylabel("Subgroup", fontdict={"fontsize": fontsize})
if show:
plt.show()
plt.close()
else:
out = outdir / f"{prefix}largest.png"
plt.gcf().set_size_inches(w=8, h=8)
plt.savefig(out, dpi=300)
plt.close()
print(f"Largest λ violin plot saved to {relpath(out)}")
def make_marchenko_plots(shifted: bool = False):
for fullpre in [True, False]:
ARGS.fullpre = fullpre
for normalize in [False]:
ARGS.normalize = normalize
datasets_all = DATASETS_FULLPRE if ARGS.fullpre else DATASETS
datasets = {}
for dataset_name, dataset in datasets_all.items():
if dataset_name == "SINGLE_SUBJECT":
continue
datasets[dataset_name] = dataset
# print(len(datasets)) # 12
# sys.exit(1)
fig: plt.Figure
fig, axes = plt.subplots(nrows=4, ncols=3)
suptitle = (
f"{'Shifted ' if shifted else ''}Marchenko Noise Ratio {'(preprocessed)' if ARGS.fullpre else ''}"
)
for i, dataset_name in enumerate(datasets):
dfs = []
for groupname, observables in precompute_dataset(dataset_name, ARGS, silent=True).items():
df_full = pd.read_pickle(observables["marchenko"])
df = pd.DataFrame(df_full.loc["noise_ratio_shifted" if shifted else "noise_ratio", :])
df["subgroup"] = [groupname for _ in range(len(df))]
dfs.append(df)
df = pd.concat(dfs)
sbn.set_context("paper")
sbn.set_style("ticks")
# args_prefix = argstrings_from_args(args)[0]
dname = dataset_name.upper()
# prefix = f"{dname}_{args_prefix}_{'fullpre_' if args.fullpre else ''}"
title = f"{dname}"
with sbn.axes_style("ticks"):
subfontsize = 10
fontsize = 12
ax: plt.Axes = axes.flat[i]
sbn.violinplot(x="subgroup", y=f"noise_ratio{'_shifted' if shifted else ''}", data=df, ax=ax)
sbn.despine(offset=10, trim=True, ax=ax)
ax.set_title(title, fontdict={"fontsize": fontsize})
ax.set_xlabel("", fontdict={"fontsize": subfontsize})
ax.set_ylabel("", fontdict={"fontsize": subfontsize})
fig.suptitle(suptitle)
fig.text(x=0.5, y=0.04, s="Subgroup", ha="center", va="center") # xlabel
fig.text(x=0.05, y=0.5, s="Noise Proportion", ha="center", va="center", rotation="vertical") # ylabel
fig.subplots_adjust(hspace=0.48, wspace=0.3)
# sbn.despine(ax=axes.flat[-1], trim=True)
fig.delaxes(ax=axes.flat[-1])
plt.show(block=False)
plt.show()
def compute_all_preds_df(args: Any,
silent: bool = False,
force: bool = False) -> Path:
preds_out, _ = stats_fnames(args, args.normalize, extension="csv")
all_preds_out = DATA_ROOT / preds_out
if not force and all_preds_out.exists():
return all_preds_out
DUDS = [
"control_pre_v_parkinsons", "park_pre_v_parkinsons",
"control_v_park_pre", "control_v_control_pre"
]
datasets = DATASETS_FULLPRE if args.fullpre else DATASETS
all_preds = []
for dataset_name, dataset in datasets.items():
if dataset_name == "SINGLE_SUBJECT":
continue
precompute_dataset(dataset_name,
args=args,
force_all=False,
silent=silent)
pairings = Pairings(args, dataset_name)
for pair in pairings.pairs:
if pair.label in DUDS:
continue
preds = pair.paired_predicts(args=args,
logistic=True,
normalize=args.normalize,
silent=silent)
all_preds.append(preds)
preds_df = pd.concat(all_preds)
preds_df.to_csv(all_preds_out)
if not silent:
print(preds_df)
print(f"Saved all predictions to {all_preds_out}")
return all_preds_out
def make_largest_plots():
for fullpre in [True, False]:
ARGS.fullpre = fullpre
for NORMALIZE in [False]:
datasets_all = DATASETS_FULLPRE if ARGS.fullpre else DATASETS
datasets = {}
for dataset_name, dataset in datasets_all.items():
if dataset_name == "SINGLE_SUBJECT":
continue
datasets[dataset_name] = dataset
# print(len(datasets)) # 12
# sys.exit(1)
fig: plt.Figure
fig, axes = plt.subplots(nrows=4, ncols=3)
suptitle = f"Largest Eigenvalue{' (preprocessed)' if ARGS.fullpre else ''}"
for i, dataset_name in enumerate(datasets):
dfs = []
for groupname, observables in precompute_dataset(dataset_name, ARGS, silent=True).items():
df_full = pd.read_pickle(observables["largest"])
df = pd.DataFrame(df_full.loc["largest", :], dtype=float)
df["subgroup"] = [groupname for _ in range(len(df))]
dfs.append(df)
df = pd.concat(dfs)
sbn.set_context("paper")
sbn.set_style("ticks")
dname = dataset_name.upper()
title = f"{dname}"
with sbn.axes_style("ticks"):
subfontsize = 10
fontsize = 12
ax: plt.Axes = axes.flat[i]
sbn.violinplot(x="subgroup", y="largest", data=df, ax=ax)
sbn.despine(offset=10, trim=True, ax=ax)
ax.set_title(title, fontdict={"fontsize": fontsize})
ax.set_xlabel("", fontdict={"fontsize": subfontsize})
ax.set_ylabel("", fontdict={"fontsize": subfontsize})
fig.suptitle(suptitle)
fig.text(x=0.5, y=0.04, s="Subgroup", ha="center", va="center") # xlabel
fig.text(x=0.05, y=0.5, s="Magnitude", ha="center", va="center", rotation="vertical") # ylabel
fig.subplots_adjust(hspace=0.48, wspace=0.3)
# sbn.despine(ax=axes.flat[-1], trim=True)
fig.delaxes(ax=axes.flat[-1])
plt.show(block=False)
plt.show()
def plot_marchenko(args: Any, show: bool = False) -> None:
"""Create violin plots of the distributions of marchenko noise percents
Parameters
----------
args: Args
Contains the unfolding, trimming, normalization, etc options defined in
run.py
show: bool
If False (default) just save the plot. Otherwise, call plt.show() and do
NOT save
"""
# label is always g1_v_g2, we want "attention" to be orange, "nonattend"
# to be black
if not show:
outdir = DATA_ROOT / "plots"
os.makedirs(outdir, exist_ok=True)
datasets = DATASETS_FULLPRE if args.fullpre else DATASETS
total = 0
fig, axes = plt.subplots(nrows=3, nocols=8)
for dataset_name in datasets:
for groupname, observables in precompute_dataset(dataset_name, args, silent=True).items():
total += 1 # count how much plots we need
fig, axes = plt.subplots(nrows=3, nocols=8)
for dataset_name in datasets:
dfs, dfs_noise = [], []
for groupname, observables in precompute_dataset(dataset_name, args, silent=True).items():
df_full = pd.read_pickle(observables["marchenko"])
df = pd.DataFrame(df_full.loc["noise_ratio", :])
df_noise = pd.DataFrame(df_full.loc["noise_ratio_shifted", :])
n = len(df)
df["subgroup"] = [groupname for _ in range(n)]
df_noise["subgroup"] = [groupname for _ in range(n)]
dfs.append(df)
dfs_noise.append(df_noise)
df = pd.concat(dfs)
df_noise = pd.concat(dfs_noise)
sbn.set_context("paper")
sbn.set_style("ticks")
args_prefix = argstrings_from_args(args)[0]
dname = dataset_name.upper()
prefix = f"{dname}_{args_prefix}_{'fullpre_' if args.fullpre else ''}"
title = f"{dname} - Marchenko Noise Ratio"
title_shifted = f"{title} (shifted)"
subtitle = args_prefix.replace("_", " ")
with sbn.axes_style("ticks"):
subfontsize = 12
fontsize = 16
ax: plt.Axes = sbn.violinplot(x="subgroup", y="noise_ratio", data=df)
sbn.despine(offset=10, trim=True)
plt.gcf().suptitle(title, fontsize=fontsize)
ax.set_title(subtitle, fontdict={"fontsize": subfontsize})
ax.set_xlabel("Noise Proportion", fontdict={"fontsize": fontsize})
ax.set_ylabel("Subgroup", fontdict={"fontsize": fontsize})
if show:
plt.show()
plt.close()
else:
out = outdir / f"{prefix}marchenko.png"
plt.gcf().set_size_inches(w=8, h=8)
plt.savefig(out)
plt.close()
print(f"Marchenko plot saved to {relpath(out)}")
ax = sbn.violinplot(x="subgroup", y="noise_ratio_shifted", data=df_noise)
sbn.despine(offset=10, trim=True)
plt.gcf().suptitle(title_shifted, fontsize=fontsize)
ax.set_title(subtitle, fontdict={"fontsize": subfontsize})
ax.set_xlabel("Noise Proportion", fontdict={"fontsize": fontsize})
ax.set_ylabel("Subgroup", fontdict={"fontsize": fontsize})
if show:
plt.show()
plt.close()
else:
out = outdir / f"{prefix}marchenko_shifted.png"
plt.gcf().set_size_inches(w=8, h=8)
plt.savefig(out)
plt.close()
print(f"Marchenko shifted plot saved to {relpath(out)}")