def run(dest, results_path, seed, version):
np.random.seed(seed)
human = pd\
.read_csv(os.path.join(results_path, "human_fall_responses.csv"))\
.set_index(['version', 'block', 'stimulus', 'kappa0'])['median']\
.sortlevel()
human_raw = pd\
.read_csv(os.path.join(results_path, "human_fall_responses_raw.csv"))\
.set_index(['version', 'block', 'stimulus', 'kappa0', 'pid'])['fall? response']\
.sortlevel()
human_version = human.ix[version]
human_exp1 = human.ix['H']
human_exp2 = human.ix['G']
model = pd.read_csv(
os.path.join(results_path, "single_model_fall_responses.csv"))
results = {}
for block in ['A', 'B']:
h1 = human_exp1.ix[block]
h2 = human_exp2.ix[block]
results[(block, 'H', 'G')] = util.bootcorr(h1, h2)
for query in model['query'].unique():
hv = human_version.ix[block]
m = model\
.groupby(['block', 'query'])\
.get_group((block, query))\
.set_index(['stimulus', 'kappa0'])['median']\
.sortlevel()\
.ix[hv.index]
results[(block, query, 'Human')] = util.bootcorr(m, hv)
for version in ['H', 'G']:
hraw = np.asarray(human_raw.ix[(version, block)].unstack('pid'))
corrs = np.array([corr(hraw) for _ in range(10000)])
results[(block, version, version)] = pd.Series(
np.percentile(corrs, [2.5, 50, 97.5]),
index=['lower', 'median', 'upper'])
results = pd.DataFrame.from_dict(results).T
results.index = pd.MultiIndex.from_tuples(results.index,
names=['block', 'X', 'Y'])
results.to_csv(dest)
def run(data, results_path, seed):
np.random.seed(seed)
means = pd.read_csv(results_path.joinpath("accuracy_means.csv"))
results = {}
exclude = ['expA', 'expB', 'gs']
for (model, flipped), df in means.groupby(['model', 'flipped']):
if model in exclude:
continue
results[(model, flipped)] = util.bootcorr(
df['modtheta'], df['median'], method='spearman')
results = pd.DataFrame.from_dict(results, orient='index')
results.index = pd.MultiIndex.from_tuples(
results.index, names=['model', 'flipped'])
pth = results_path.joinpath(filename)
results.to_csv(pth)
with open(results_path.joinpath(texname), "w") as fh:
fh.write("%% AUTOMATICALLY GENERATED -- DO NOT EDIT!\n")
for (model, flipped), stats in results.iterrows():
cmd = util.newcommand(
"%sThetaAccuracyCorr%s" % (
model.capitalize(),
flipped.capitalize()),
util.latex_spearman.format(**dict(stats)))
fh.write(cmd)
return pth
def run(data, results_path, seed):
np.random.seed(seed)
keys = ['exp', 'expA', 'expB']
means = pd.read_csv(results_path.joinpath("trial_accuracy_means.csv"))
results = {}
for key in keys:
df = means.groupby('model').get_group(key)
trials = df['trial']
accuracy = df['median']
corr = util.bootcorr(trials, accuracy, method='spearman')
results[key] = corr
results = pd.DataFrame.from_dict(results, orient='index')
results.index.name = 'model'
pth = results_path.joinpath(filename)
results.to_csv(pth)
with open(results_path.joinpath(texname), "w") as fh:
fh.write("%% AUTOMATICALLY GENERATED -- DO NOT EDIT!\n")
for model, stats in results.iterrows():
cmd = util.newcommand(
"%sTrialAccuracyCorr" % model.capitalize(),
util.latex_spearman.format(**dict(stats)))
fh.write(cmd)
return pth
def run(data, results_path, seed):
np.random.seed(seed)
means = pd.read_csv(results_path.joinpath("response_time_means.csv"))
means = means\
.set_index(['stimulus', 'theta', 'flipped', 'model'])['median']\
.unstack('model')
results = {}
exclude = ['exp', 'expA', 'expB', 'gs']
for key in means:
if key in exclude:
continue
for flipped in ['same', 'flipped']:
corr = util.bootcorr(
means['exp'].unstack('flipped')[flipped],
means[key].unstack('flipped')[flipped],
method='pearson')
results[(key, flipped)] = corr
corr = util.bootcorr(
means['exp'],
means[key],
method='pearson')
results[(key, 'all')] = corr
results = pd.DataFrame.from_dict(results, orient='index')
results.index = pd.MultiIndex.from_tuples(
results.index, names=['model', 'flipped'])
pth = results_path.joinpath(filename)
results.to_csv(pth)
with open(results_path.joinpath(texname), "w") as fh:
fh.write("%% AUTOMATICALLY GENERATED -- DO NOT EDIT!\n")
for (model, flipped), stats in results.iterrows():
if flipped != 'all':
continue
cmd = util.newcommand(
"%sTimeCorr" % model.capitalize(),
util.latex_pearson.format(**dict(stats)))
fh.write(cmd)
return pth
def bootcorr(x, y, **kwargs):
name, x = x
corr = util.bootcorr(x, y, **kwargs)
corr.name = name
return corr
