def format_scores_report(table):
output = pd.DataFrame()
for column in table.columns:
output[column] = list(format_scores_column(table[column]))
output.index = table.index
output.columns = pd.MultiIndex.from_tuples(output.columns)
return table_html(output)
def format_report(table, labels, github=False):
models = table.model.unique()
datasets = table.dataset.unique()
metrics = ['errors', 'prec', 'recall', 'time']
table['errors'] = table.prec + table.recall
table = table.pivot('model', 'dataset', metrics)
table = table.swaplevel(axis=1)
if github:
metrics = ['errors', 'time']
table = table.reindex(
index=models,
columns=[
(dataset, metric)
for dataset in datasets
for metric in metrics
]
)
for dataset in datasets:
for metric in metrics:
column = table[dataset, metric]
table[dataset, metric] = list(format_column(column, metric, github))
table.index.name = None
table.columns.names = None, None
table.index = [labels.get(_, _) for _ in table.index]
return table_html(table)
def format_natasha_report(scores, bench, models):
scores = scores.loc[models]
bench = bench.loc[models]
scores.pop((GAREEV, LOC))
for column in scores.columns:
scores[column] = scores[column].map(scores_f1)
scores = scores.stack().mean(axis=1).unstack() # model x type
output = pd.DataFrame()
output['PER/LOC/ORG f1'] = [
format_f1_scores(_)
for _ in scores[[PER, LOC, ORG]].values
]
columns = [
['init', format_sec, 'init, s'],
['disk', format_mb, 'disk, mb'],
['ram', format_mb, 'ram, mb'],
[['speed', 'device'], format_speed, 'speed, articles/s']
]
for slice, format, name in columns:
values = bench[slice].values.tolist()
output[name] = [format(_) for _ in values]
output.index = models
return table_html(output)
def format_natasha_report(table, labels, models):
table = table.groupby('model')['errors', 'time'].sum()
table = table.loc[models]
table.index.name = None
table.index = [labels.get(_, _) for _ in table.index]
return table_html(table)
def format_scores_report(table):
output = pd.DataFrame()
for column in table.columns:
output[column] = table[column].map(format_scores)
output.columns = pd.MultiIndex.from_tuples(output.columns)
output.columns.names = [None, 'prec/recall/f1,%']
return table_html(output)
def format_github_report2(table, datasets):
output = pd.DataFrame()
output['type'] = table['type']
for dataset in datasets:
values = table[dataset].values
selection = select_max(values, key=first)
values = highlight(values, selection, format_github_cell)
output[dataset] = list(values)
output = output.rename(columns={'simlex965': 'simlex'})
return table_html(output)
def format_github_scores_report(table):
output = pd.DataFrame()
for column in table.columns:
dataset, type = column
if dataset == GAREEV and type == LOC:
continue
output[column] = list(format_github_scores_column(table[column]))
output.index = table.index
output.columns = pd.MultiIndex.from_tuples(output.columns)
output.columns.names = [None, 'f1']
return table_html(output)
def format_github_report1(table):
output = pd.DataFrame()
output['type'] = table['type']
columns = [['init', format_sec, select_min, 'init, s'],
['get', format_mks, select_min, 'get, µs'],
['disk', format_mb, select_min, 'disk, mb'],
['ram', format_mb, select_min, 'ram, mb'],
['vocab', format_vocab, select_max, 'vocab']]
for column, format, select, name in columns:
values = table[column].values
selection = select(values)
values = highlight(values, selection, format)
output[name] = list(values)
return table_html(output)
def format_report(table, datasets):
output = pd.DataFrame()
output['type'] = table['type']
columns = [
['init', format_sec, 'init, s'],
['get', format_mks, 'get, µs'],
['disk', format_mb, 'disk, mb'],
['ram', format_mb, 'ram, mb'],
]
for column, format, name in columns:
output[name] = table[column].map(format)
for dataset in datasets:
output[dataset] = table[dataset].map(format_cell)
return table_html(output)
def format_natasha_report(scores, bench, dataset, models):
scores = scores.loc[models]
bench = bench.loc[models]
output = pd.DataFrame()
output['accuracy'] = [
'%.3f' % _
for _ in scores[dataset]
]
columns = [
['init', format_sec, 'init, s'],
['disk', format_mb, 'disk, mb'],
['ram', format_mb, 'ram, mb'],
[['speed', 'device'], format_speed, 'speed, sents/s']
]
for slice, format, name in columns:
values = bench[slice].values.tolist()
output[name] = [format(_) for _ in values]
output.index = models
return table_html(output)
def format_natasha_report(table, datasets):
output = pd.DataFrame()
output['type'] = table['type']
scores = pd.DataFrame()
for dataset in datasets:
scores[dataset] = [score for score, cover in table[dataset]]
scores = scores.mean(axis=1)
output['precision'] = ['%.3f' % _ for _ in scores]
columns = [
['init', format_sec, 'init, s'],
['disk', format_mb, 'disk, mb'],
['ram', format_mb, 'ram, mb'],
['vocab', format_vocab, 'vocab'],
]
for column, format, name in columns:
values = table[column].values
output[name] = [format(_) for _ in values]
output.index = table.index
return table_html(output)
def format_scores_report(table):
output = pd.DataFrame()
for column in table.columns:
output[column] = list(format_scores_column(table[column]))
output.index = table.index
return table_html(output)
