def main(dataset: str, target: str, pipeline: str):
shapes = []
ds_service = DatasetService()
m_service = ModelService()
for symbol in SYMBOLS:
print(f"Exporting shap dataframes for symbol {symbol}")
ds = ds_service.get_dataset(name=dataset, symbol=symbol)
fs = DatasetService.get_feature_selection(ds=ds,
method='importances_shap',
target=target)
X_all = ds_service.get_dataset_features(ds=ds, columns=fs.features)
y_all = ds_service.get_dataset_target(ds=ds, name=target)
model = m_service.get_model(pipeline=pipeline,
dataset=dataset,
target=target,
symbol=symbol)
for t in model.tests:
print(f"Loading estimators for test {t.window}")
estimators = ModelService.load_test_estimators(model=model, mt=t)
shaps = []
print(f"Calculating shap values...")
for est in tqdm(estimators):
est_class = y_all.loc[est.day]
shap_v, shap_exp = get_shap_values(estimator=est,
X=X_all.loc[est.day],
X_train=est.train_x,
bytes=False)
df = pd.DataFrame([shap_v],
index=[pd.to_datetime(est.day)],
columns=X_all.columns)
df['label'] = y_all.loc[est.day]
df['shap_expected'] = shap_exp
shaps.append(df)
print("Exporting dataframe..")
cdf = pd.concat(shaps, axis='index')
os.makedirs(f"data/shap_values/{dataset}/{target}/{pipeline}/",
exist_ok=True)
cdf.to_csv(
f"data/shap_values/{dataset}/{target}/{pipeline}/shap_test_{symbol}_Wdays{t.window['days']}.csv",
index_label='time')
print("Exported.")
# # Load day estimator
# est = load_estimator()
print(f"Plotted {symbol}")
def main(dataset: str):
dss = DatasetService()
records = []
for symbol in SYMBOLS:
ds = dss.get_dataset(name=dataset, symbol=symbol)
fs = DatasetService.get_feature_selection(ds, 'importances_shap',
'class')
target = dss.get_dataset_target(ds=ds, name='class')
uniq, cnt = np.unique(target, return_counts=True)
if cnt[0] + cnt[1] + cnt[2] != ds.count:
print(f"Mismatch between classes and count in {symbol}")
mindt = from_timestamp(ds.valid_index_min)
maxdt = from_timestamp(ds.valid_index_max)
daysn = (maxdt - mindt).days
records.append({
'Pair': symbol,
'num_features': len(ds.features),
'sel_features': len(fs.features),
'min_index': ds.valid_index_min,
'max_index': ds.valid_index_max,
'valid_days': daysn,
'records': ds.count,
'sell_count': cnt[0],
'hold_count': cnt[1],
'buy_count': cnt[2]
})
df = pd.DataFrame.from_records(records)
fig = px.timeline(df, x_start="min_index", x_end="max_index", y="Pair")
fig.update_yaxes(
autorange="reversed") # otherwise tasks are listed from the bottom up
#fig.show()
fig.update_layout(title={
'text': f"Sample distribution across datasets",
'y': 0.9,
'x': 0.5,
'xanchor': 'center',
'yanchor': 'top'
},
paper_bgcolor='rgba(0,0,0,0)',
plot_bgcolor='rgba(0,0,0,0.5)',
font={'color': 'White'},
margin={
'l': 5,
'r': 5,
't': 80,
'b': 5,
'pad': 5
})
fig.write_image("images/data_summary/timeline.png")
for symbol in SYMBOLS:
sdf = df[df.Pair == symbol]
pie_values = [
sdf['sell_count'].values[0], sdf['hold_count'].values[0],
sdf['buy_count'].values[0]
]
pie_labels = ['SELL', 'HOLD', 'BUY']
sfig = go.Figure(data=[
go.Pie(
labels=pie_labels,
values=pie_values,
textinfo='label+percent',
#insidetextorientation='radial',
showlegend=False)
])
sfig.update_layout(title={
'text': f"Class distribution for pair {symbol}",
'y': 0.9,
'x': 0.5,
'xanchor': 'center',
'yanchor': 'top',
'font': {
'size': 22
}
},
paper_bgcolor='rgba(0,0,0,0)',
plot_bgcolor='rgba(0,0,0,0)',
font={
'color': 'White',
'size': 26
},
margin={
'l': 0,
'r': 0,
't': 80,
'b': 0,
'pad': 0
},
uniformtext_minsize=24)
sfig.write_image(f"images/data_summary/{symbol}_distribution.png")
print(df.head())
def main(dataset: str, target: str):
num_shap_plots = 3
shap_show_count = 10
ds_service = DatasetService()
m_service = ModelService()
for pipeline in PIPELINES:
for symbol in SYMBOLS:
print(
f"Plotting shap dataframes for pipeline {pipeline} symbol {symbol}"
)
ds = ds_service.get_dataset(name=dataset, symbol=symbol)
fs = DatasetService.get_feature_selection(
ds=ds, method='importances_shap', target=target)
X_all = ds_service.get_dataset_features(ds=ds, columns=fs.features)
y_all = ds_service.get_dataset_target(ds=ds, name=target)
model = m_service.get_model(pipeline=pipeline,
dataset=dataset,
target=target,
symbol=symbol)
for t in model.tests:
placeholder = "{label}"
csv_name = f"data/shap_values/{dataset}/{target}/{pipeline}/shap_training_window_{symbol}_{placeholder}_Wdays{t.window['days']}_.csv"
expected_csv_name = csv_name.format(label='SHAP_expected')
print(f"Loading results for test {t.window}")
results = ModelService.parse_test_results(test=t)
exp_shap_df = pd.read_csv(expected_csv_name,
index_col='time',
parse_dates=True)
for cls, label in enumerate(["SELL", "HOLD", "BUY"]):
class_csv_name = csv_name.format(label=label)
cls_shap_df = pd.read_csv(class_csv_name,
index_col='time',
parse_dates=True)
cls_shap_df = cls_shap_df.loc[t.test_interval.begin:t.
test_interval.end]
x_train = X_all.loc[cls_shap_df.index]
chunk_size = int(cls_shap_df.shape[0] / num_shap_plots)
fig = plt.figure(constrained_layout=True,
figsize=(100, 50),
dpi=300) #
gs = GridSpec(3,
num_shap_plots,
figure=fig,
wspace=1.5,
hspace=0.3)
precision_ax = fig.add_subplot(gs[0, :])
shap_values_ax = fig.add_subplot(gs[1, :])
beeswarms_axs = [
fig.add_subplot(gs[2, i])
for i in range(num_shap_plots)
]
#format_axes(fig)
shap_plot_labels = set()
first_shap_day = results.iloc[0]['time'].replace(
'+00:00',
'').replace('T', '').replace(':', '').replace('-', '')
middle_shap_day = results.iloc[int(
results.shape[0] / 2)]['time'].replace(
'+00:00',
'').replace('T', '').replace(':',
'').replace('-', '')
last_shap_day = results.iloc[-1]['time'].replace(
'+00:00',
'').replace('T', '').replace(':', '').replace('-', '')
for idx, dayname in enumerate(
[first_shap_day, middle_shap_day, last_shap_day]):
day_csv_name = f"data/shap_values/{dataset}/{target}/{pipeline}/daily/shap_training_window_{symbol}_{label}_Wdays{t.window['days']}_DAY{dayname}.csv"
# Plot each section's SHAP values
cdf_subset = pd.read_csv(day_csv_name,
index_col='time',
parse_dates=True)
train_subset = X_all.loc[cdf_subset.index]
# Get a rank of feature labels based on this section's shap values
abs_mean_shap = cdf_subset.abs().mean(axis='index')
abs_mean_rank = abs_mean_shap.sort_values(
ascending=False)[:shap_show_count]
for l in abs_mean_rank.index:
# Save labels for features in the top-N
shap_plot_labels.add(l)
# Plot this section's SHAP values
plt.sca(beeswarms_axs[idx])
shap.summary_plot(cdf_subset.values,
train_subset,
max_display=shap_show_count,
show=False,
color_bar=False,
sort=True)
min_date = cdf_subset.index.min().to_pydatetime()
max_date = cdf_subset.index.max().to_pydatetime(
) + timedelta(days=1)
min_date_f = min_date.strftime("%Y/%m/%d")
max_date_f = max_date.strftime("%Y/%m/%d")
beeswarms_axs[idx].set_xlabel(
f"SHAP values\nWindow: {min_date_f} - {max_date_f}",
fontsize=8)
beeswarms_axs[idx].tick_params(axis='y',
which='major',
labelsize=6)
beeswarms_axs[idx].tick_params(axis='x',
which='major',
labelsize=8)
# Plot shap values
day_csv_name = f"data/shap_values/{dataset}/{target}/{pipeline}/shap_training_window_{symbol}_{label}_Wdays{t.window['days']}_.csv"
plot_cls_shap_df = pd.read_csv(day_csv_name,
index_col='time',
parse_dates=True)
def get_spread(series):
return np.abs(series.max() - series.min())
plot_rank = plot_cls_shap_df[list(shap_plot_labels)].apply(
get_spread, axis='index').sort_values(
ascending=False)[:shap_show_count]
plot_cls_shap_df['xlabel'] = [
t.to_pydatetime().strftime("%Y/%m/%d")
for t in plot_cls_shap_df.index
]
shap_ax = plot_cls_shap_df.plot(
x='xlabel',
y=[c for c in plot_rank.index],
kind='line',
ax=shap_values_ax,
legend=False,
xlabel='')
patches, labels = shap_ax.get_legend_handles_labels()
shap_ax.legend(patches,
labels,
loc='center left',
bbox_to_anchor=(1, 0.5),
prop={'size': 6})
shap_ax.tick_params(axis='x', which='major', labelsize=8)
shap_ax.set_ylabel('mean(|SHAP|)', fontsize=6)
#shap_ax.tick_params(labelbottom=False, labelleft=False)
# Get Metrics scores dataframe
cri_df = get_metrics_df(results).rolling(
7, min_periods=1).mean()
cri_df['xlabel'] = [
t.to_pydatetime().strftime("%Y/%m/%d")
for t in cri_df.index
]
cri_ax = cri_df.plot(x='xlabel',
y=f"pre_{cls}",
kind='line',
ax=precision_ax,
legend=False,
xlabel='')
patches, labels = cri_ax.get_legend_handles_labels()
cri_ax.legend(patches,
labels,
loc='center left',
bbox_to_anchor=(1, 0.5),
prop={'size': 6})
cri_ax.set_ylabel('mean(precision)', fontsize=6)
cri_ax.tick_params(labelbottom=False, labelleft=True)
min_date = cri_df.index.min().to_pydatetime().strftime(
"%Y/%m/%d")
max_date = cri_df.index.max().to_pydatetime().strftime(
"%Y/%m/%d")
window = t.window['days']
fig.suptitle(
f"{symbol}, {pipeline}, W={window}D, Class {label}, From {min_date} to {max_date}"
)
# fig.show()
os.makedirs(f"images/shap-test-final/", exist_ok=True)
plt.savefig(
f"images/shap-test-final/{pipeline}_W{window}D_{dataset}_{target}_{symbol}_{label}.png",
dpi='figure')
plt.close()
print(f"{label} OK")
print(f"Exported symbol {symbol}.")
# # Load day estimator
# est = load_estimator()
print(f"Plotted {symbol}")
def main(dataset: str, target: str, pipeline: str):
shapes = []
ds_service = DatasetService()
m_service = ModelService()
for symbol in SYMBOLS:
print(f"Exporting shap dataframes for symbol {symbol}")
ds = ds_service.get_dataset(name=dataset, symbol=symbol)
fs = DatasetService.get_feature_selection(ds=ds,
method='importances_shap',
target=target)
X_all = ds_service.get_dataset_features(ds=ds, columns=fs.features)
y_all = ds_service.get_dataset_target(ds=ds, name=target)
model = m_service.get_model(pipeline=pipeline,
dataset=dataset,
target=target,
symbol=symbol)
for t in model.tests:
os.makedirs(
f"data/shap_values/{dataset}/{target}/{pipeline}/daily",
exist_ok=True)
placeholder = "{label}"
csv_name = f"data/shap_values/{dataset}/{target}/{pipeline}/shap_training_window_{symbol}_{placeholder}_Wdays{t.window['days']}_.csv"
day_csv_name = f"data/shap_values/{dataset}/{target}/{pipeline}/daily/shap_training_window_{symbol}_{placeholder}_Wdays{t.window['days']}_"
print(f"Loading estimators for test {t.window}")
estimators = ModelService.load_test_estimators(model=model, mt=t)
results = ModelService.parse_test_results(test=t)
shaps = [[], [], []]
X_test = X_all.loc[t.test_interval.begin:t.test_interval.end]
shap_expected = []
print(f"Calculating shap values")
shap_abs_mean = [pd.DataFrame(), pd.DataFrame(), pd.DataFrame()]
for est in tqdm(estimators):
est_class = y_all.loc[est.day]
training_data = est.train_x.astype(np.float64).fillna(value=0)
shap_v, shap_exp = get_shap_values(estimator=est.named_steps.c,
X=training_data,
X_train=training_data,
bytes=False)
if isinstance(shap_exp, float):
shap_expected.append([est.day] + [0, 0, shap_exp])
else:
shap_expected.append([est.day] + [v for v in shap_exp])
for cls, label in enumerate(["SELL", "HOLD", "BUY"]):
df = pd.DataFrame(shap_v[cls],
index=est.train_x.index,
columns=est.train_x.columns)
# if not shaps[cls]: # If list is empty, append whole df
# shaps[cls].append(df)
# else:
# shaps[cls].append(df.iloc[-1:]) # otherwise only append new row (sliding window)
# Save shap values dataframe for each day
dayname = est.day.replace('+00:00',
'').replace('T', '').replace(
':', '').replace('-', '')
day_class_csv_name = day_csv_name.format(
label=label) + f"DAY{dayname}.csv"
df.to_csv(day_class_csv_name, index_label='time')
# Process data for next plot
df_abs_mean = df.abs().mean().to_dict()
df_abs_mean['time'] = est.day
shaps[cls].append(df_abs_mean)
# print(shap_abs_mean.head())
# Merge shap values in an unique dataframe and save to csv for each class
for cls, label in enumerate(["SELL", "HOLD", "BUY"]):
class_csv_name = csv_name.format(label=label)
print(
f"Exporting dataframe for class {label} -> {class_csv_name}"
)
# cdf = pd.concat(shaps[cls], axis='index')
cdf = pd.DataFrame.from_records(shaps[cls])
cdf.index = pd.to_datetime(cdf.time)
cdf = cdf[cdf.columns.difference(['time'])]
cdf.to_csv(class_csv_name, index_label='time')
expected_csv_name = csv_name.format(label='SHAP_expected')
print(
f"Exporting expected values dataframe -> {expected_csv_name}")
edf = pd.DataFrame(
shap_expected,
columns=[
"time", "shap_expected_sell", "shap_expected_hold",
"shap_expected_buy"
],
)
edf.to_csv(expected_csv_name, index_label='time')
print(f"Exported symbol {symbol}.")
# # Load day estimator
# est = load_estimator()
print(f"Plotted {symbol}")
def main(dataset: str, target: str, pipeline: str):
hierarchy = load_hierarchy(f"{dataset}_{target}_feature_hierarchy.yml")
hdf = pd.DataFrame(hierarchy)
num_shap_plots = 3
shap_show_count = 10
ds_service = DatasetService()
m_service = ModelService()
for symbol in SYMBOLS:
print(f"Plotting shap dataframes for symbol {symbol}")
ds = ds_service.get_dataset(name=dataset, symbol=symbol)
fs = DatasetService.get_feature_selection(ds=ds,
method='importances_shap',
target=target)
X_all = ds_service.get_dataset_features(ds=ds, columns=fs.features)
y_all = ds_service.get_dataset_target(ds=ds, name=target)
model = m_service.get_model(pipeline=pipeline,
dataset=dataset,
target=target,
symbol=symbol)
for t in model.tests:
os.makedirs(
f"images/shap-test-hierarchy/{dataset}/{target}/{pipeline}/",
exist_ok=True)
placeholder = "{label}"
csv_name = f"data/shap_values/{dataset}/{target}/{pipeline}/shap_training_window_{symbol}_{placeholder}_Wdays{t.window['days']}_.csv"
expected_csv_name = csv_name.format(label='SHAP_expected')
print(f"Loading results for test {t.window}")
results = ModelService.parse_test_results(test=t)
exp_shap_df = pd.read_csv(expected_csv_name,
index_col='time',
parse_dates=True)
for cls, label in enumerate(["SELL", "HOLD", "BUY"]):
class_csv_name = csv_name.format(label=label)
cls_shap_df = pd.read_csv(class_csv_name,
index_col='time',
parse_dates=True)
cls_shap_df = cls_shap_df.loc[t.test_interval.begin:t.
test_interval.end]
x_train = X_all.loc[cls_shap_df.index]
chunk_size = int(cls_shap_df.shape[0] / num_shap_plots)
# fig = plt.figure(constrained_layout=True, figsize=(100, 50), dpi=300) #
# gs = GridSpec(3, num_shap_plots, figure=fig, wspace=1.5, hspace=0.3)
# precision_ax = fig.add_subplot(gs[0, :])
# shap_values_ax = fig.add_subplot(gs[1, :])
# beeswarms_axs = [fig.add_subplot(gs[2, i]) for i in range(num_shap_plots)]
# #format_axes(fig)
# shap_plot_labels = set()
# for idx, start in enumerate(range(0, cls_shap_df.shape[0], chunk_size)):
# end = start + chunk_size
# left = cls_shap_df.shape[0] - end
# if left > 0 and left < chunk_size:
# end += left
# elif left < 0:
# break
# # Plot each section's SHAP values
# cdf_subset = cls_shap_df.iloc[start:end]
# train_subset = x_train.iloc[start:end]
#
# # Get a rank of feature labels based on this section's shap values
# abs_mean_shap = cdf_subset.abs().mean(axis='index')
# abs_mean_rank = abs_mean_shap.sort_values(ascending=False)[:shap_show_count]
# for l in abs_mean_rank.index:
# # Save labels for features in the top-N
# shap_plot_labels.add(l)
#
# # Plot this section's SHAP values
# plt.sca(beeswarms_axs[idx])
# shap.summary_plot(
# cdf_subset.values,
# train_subset,
# max_display=shap_show_count,
# show=False,
# color_bar=False,
# sort=True
# )
# min_date = cdf_subset.index.min().to_pydatetime().strftime("%Y/%m/%d")
# max_date = cdf_subset.index.max().to_pydatetime().strftime("%Y/%m/%d")
# beeswarms_axs[idx].set_xlabel(f"SHAP values\n{min_date} - {max_date}", fontsize=8)
# beeswarms_axs[idx].tick_params(axis='y', which='major', labelsize=6)
# beeswarms_axs[idx].tick_params(axis='x', which='major', labelsize=8)
# # Plot shap values
# plot_cls_shap_df = cls_shap_df.abs().rolling(7, min_periods=1).mean()
# def get_spread(series):
# return np.abs(series.max() - series.min())
# plot_rank = plot_cls_shap_df[list(shap_plot_labels)].apply(get_spread, axis='index').sort_values(ascending=False)[:shap_show_count]
# plot_cls_shap_df['xlabel'] = [t.to_pydatetime().strftime("%Y/%m/%d") for t in plot_cls_shap_df.index]
# shap_ax = plot_cls_shap_df.plot(
# x='xlabel',
# y=[c for c in plot_rank.index],
# kind='line',
# ax=shap_values_ax,
# legend=False,
# xlabel=''
# )
# patches, labels = shap_ax.get_legend_handles_labels()
# shap_ax.legend(
# patches, labels,
# loc='center left', bbox_to_anchor=(1, 0.5), prop={'size': 6}
# )
# shap_ax.tick_params(axis='x', which='major', labelsize=8)
# shap_ax.set_ylabel('mean(|SHAP|)', fontsize=6)
# #shap_ax.tick_params(labelbottom=False, labelleft=False)
#
# # Get Metrics scores dataframe
# cri_df = get_metrics_df(results).rolling(7, min_periods=1).mean()
# cri_df['xlabel'] = [t.to_pydatetime().strftime("%Y/%m/%d") for t in cri_df.index]
# cri_ax = cri_df.plot(
# x='xlabel',
# y=f"pre_{cls}",
# kind='line',
# ax=precision_ax,
# legend=False,
# xlabel=''
# )
# patches, labels = cri_ax.get_legend_handles_labels()
# cri_ax.legend(
# patches, labels,
# loc='center left', bbox_to_anchor=(1, 0.5), prop={'size': 6}
# )
# cri_ax.set_ylabel('mean(precision)', fontsize=6)
# cri_ax.tick_params(labelbottom=False, labelleft=True)
#
# min_date = cri_df.index.min().to_pydatetime().strftime("%Y/%m/%d")
# max_date = cri_df.index.max().to_pydatetime().strftime("%Y/%m/%d")
# fig.suptitle(f"{pipeline}, {symbol}, class {label} tests from {min_date} to {max_date}")
#
# # fig.show()
# plt.savefig(
# f"images/shap-test/{pipeline}_{dataset}_{target}_{symbol}_{label}.png",
# dpi='figure'
# )
# plt.close()
print(f"{label} OK")
print(f"Exported symbol {symbol}.")
# # Load day estimator
# est = load_estimator()
print(f"Plotted {symbol}")
class FeatureSelectionService:
def __init__(self):
self.model_repo = ModelRepository()
self.dataset_service = DatasetService()
def create_features_search(self,
*,
symbol: str,
dataset: str,
target: str,
split: float,
method: str,
task_key: str = None) -> ModelFeatures:
ds = self.dataset_service.get_dataset(dataset, symbol)
splits = DatasetService.get_train_test_split_indices(ds, split)
result = ModelFeatures(dataset=dataset,
target=target,
symbol=symbol,
search_interval=splits['train'],
feature_selection_method=method,
task_key=task_key or str(uuid4()))
return result
def feature_selection(self, mf: ModelFeatures, **kwargs) -> ModelFeatures:
# Load dataset
X = self.dataset_service.get_features(mf.dataset,
mf.symbol,
mf.search_interval.begin,
mf.search_interval.end,
columns=mf.features)
y = self.dataset_service.get_target(mf.target, mf.symbol,
mf.search_interval.begin,
mf.search_interval.end)
unique, counts = np.unique(y, return_counts=True)
if len(unique) < 2:
logging.error(
"[{}-{}-{}]Training data contains less than 2 classes: {}".
format(mf.symbol, mf.dataset, mf.target, unique))
raise MessageException(
"Training data contains less than 2 classes: {}".format(
unique))
# Perform search
mf.start_at = get_timestamp() # Log starting timestamp
if not mf.feature_selection_method or mf.feature_selection_method == 'importances':
selector = select_from_model(X, y)
mf.feature_importances = label_feature_importances(
selector.estimator_, X.columns)
elif mf.feature_selection_method == 'importances_cv':
selector = select_from_model_cv(X, y)
mf.feature_importances = label_feature_importances(
selector.estimator_.best_estimator_, X.columns)
elif mf.feature_selection_method == 'fscore':
selector = select_percentile(X, y, percentile=10)
elif mf.feature_selection_method == 'relieff':
selector = select_relieff(X, y, percentile=10)
elif mf.feature_selection_method == 'multisurf':
selector = select_multisurf(X, y, percentile=10)
else:
raise NotFoundException(
"Cannot find feature selection method by {}".format(
mf.feature_selection_method))
mf.end_at = get_timestamp() # Log ending timestamp
# Update search request with results
mf.features = label_support(selector.get_support(), X.columns)
# Update model with the new results
if kwargs.get('save', True):
self.model_repo.append_features_query(
{
"dataset": mf.dataset,
"symbol": mf.symbol,
"target": mf.target
}, mf)
return mf
def get_available_symbols(self, dataset: str):
return self.dataset_service.get_dataset_symbols(name=dataset)
def feature_selection_new(self, *, symbol: str, dataset: str, target: str,
split: float, method: str,
**kwargs) -> ModelFeatures:
ds = self.dataset_service.get_dataset(dataset, symbol)
fs_exists = DatasetService.has_feature_selection(ds=ds,
method=method,
target=target)
if fs_exists:
if kwargs.get('replace'):
self.dataset_service.remove_feature_selection(ds=ds,
method=method,
target=target)
else:
if kwargs.get('save'):
raise MessageException(
f"Feature selection with method '{method}' alrady performed for '{dataset}.{symbol}' and target '{target}'"
)
splits = DatasetService.get_train_test_split_indices(ds, split)
fs = FeatureSelection(target=target,
method=method,
search_interval=splits['train'],
task_key=kwargs.get('task_key', str(uuid4())))
# Load dataset
X = self.dataset_service.get_dataset_features(
ds=ds, begin=fs.search_interval.begin, end=fs.search_interval.end)
y = self.dataset_service.get_dataset_target(
name=fs.target,
ds=ds,
begin=fs.search_interval.begin,
end=fs.search_interval.end)
unique, counts = np.unique(y, return_counts=True)
if len(unique) < 2:
logging.error(
"[{}-{}-{}]Training data contains less than 2 classes: {}".
format(symbol, dataset, target, unique))
raise MessageException(
"Training data contains less than 2 classes: {}".format(
unique))
# Perform search
fs.start_at = get_timestamp() # Log starting timestamp
if not fs.method or 'importances' in fs.method:
if '_cv' in fs.method:
selector = select_from_model_cv(X, y)
else:
selector = select_from_model(X, y)
fs.feature_importances = label_feature_importances(
selector.estimator_, X.columns)
if '_shap' in fs.method:
fs.shap_values = get_shap_values(
model=selector.estimator_.named_steps.c, X=X, X_train=X)
shap_values = parse_shap_values(fs.shap_values)
elif fs.method == 'fscore':
selector = select_percentile(X, y, percentile=10)
elif fs.method == 'relieff':
selector = select_relieff(X, y, percentile=10)
elif fs.method == 'multisurf':
selector = select_multisurf(X, y, percentile=10)
else:
raise NotFoundException(
"Cannot find feature selection method by {}".format(fs.method))
fs.end_at = get_timestamp() # Log ending timestamp
# Update search request with results
fs.features = label_support(selector.get_support(), X.columns)
if not kwargs.get('save'):
return fs
return self.dataset_service.append_feature_selection(ds, fs)