def explain(self):
answers = {}
for pred_col in self.evaluations:
answers[pred_col] = []
prediction_row = {col: self.data[col][self.row_index] for col in list(self.data.keys())}
evaluation = self.evaluations[pred_col][self.row_index]
clusters = evaluation.explain(self.transaction_output.transaction.lmd['column_stats'][pred_col])
for cluster in clusters:
pct_confidence = round(cluster['confidence'] * 100)
predicted_value = cluster['value']
if self.col_stats[pred_col]['data_type'] in (DATA_TYPES.NUMERIC, DATA_TYPES.DATE):
value_range = [cluster['buckets'][0],cluster['buckets'][-1]]
range_pretty_start = value_range[0]
if range_pretty_start > 1000:
range_pretty_start = round(range_pretty_start)
if range_pretty_start > 100:
range_pretty_start = round(range_pretty_start,2)
if range_pretty_start > 10:
range_pretty_start = round(range_pretty_start,1)
elif range_pretty_start > pow(10,-3):
range_pretty_start = round(range_pretty_start,6)
range_end_start = value_range[-1]
if range_end_start > 1000:
range_end_start = round(range_end_start)
if range_end_start > 100:
range_end_start = round(range_end_start,2)
if range_end_start > 10:
range_end_start = round(range_end_start,1)
elif range_end_start > pow(10,-3):
range_end_start = round(range_end_start,6)
explaination = explain_prediction(self.transaction_output.transaction.lmd, prediction_row, cluster['confidence'], pred_col)
answers[pred_col].append({
'value': predicted_value,
'range': value_range,
'confidence': cluster['confidence'],
'explaination': explaination,
'simple': f'We are {pct_confidence}% confident the value of "{pred_col}" lies between {range_pretty_start} and {range_end_start}'
})
else:
explaination = explain_prediction(self.transaction_output.transaction.lmd, prediction_row, cluster['confidence'], pred_col)
answers[pred_col].append({
'value': predicted_value,
'confidence': cluster['middle_confidence'],
'explaination': explaination,
'simple': f'We are {pct_confidence}% confident the value of "{pred_col}" is {predicted_value}'
})
answers[pred_col] = sorted(answers[pred_col], key=lambda x: x['confidence'], reverse=True)
return answers
def explain(self):
prediction_row = {
col: self.transaction_output.data[col][self.row_index]
for col in list(self.transaction_output.data.keys())
}
#self.transaction_output.data.iloc[self.row_index]
return explain_prediction(self.transaction_output.transaction.lmd,
prediction_row)
def explain(self):
answers = {}
for pred_col in self.predict_columns:
answers[pred_col] = []
prediction_row = {
col: self.data[col][self.row_index]
for col in list(self.data.keys())
}
clusters = [{
'value': prediction_row[pred_col],
'confidence': prediction_row[f'{pred_col}_confidence']
}]
for cluster in clusters:
pct_confidence = round(cluster['confidence'] * 100)
predicted_value = cluster['value']
if f'{pred_col}_model_confidence' in prediction_row:
new_conf = round(
(prediction_row[f'{pred_col}_model_confidence'] * 3 +
cluster['confidence'] * 1) / 4, 4)
else:
new_conf = cluster['confidence']
explanation = explain_prediction(
self.transaction_output.transaction.lmd, prediction_row,
cluster['confidence'], pred_col)
answers[pred_col].append({
'value':
predicted_value,
'confidence':
new_conf,
'explanation':
explanation,
'explaination':
explanation,
'simple':
f'We are {pct_confidence}% confident the value of "{pred_col}" is {predicted_value}'
})
if self.transaction_output.input_confidence is not None:
for i in range(len(answers[pred_col])):
answers[pred_col][i]['confidence_influence_scores'] = {
'confidence_variation_score': [],
'column_names': []
}
for c in self.transaction_output.input_confidence:
answers[pred_col][i][
'confidence_influence_scores'][
'confidence_variation_score'].append(
self.transaction_output.
input_confidence[c])
answers[pred_col][i][
'confidence_influence_scores'][
'column_names'].append(str(c))
model_result = {'value': prediction_row[f'model_{pred_col}']}
if f'{pred_col}_model_confidence' in prediction_row:
model_result['confidence'] = prediction_row[
f'{pred_col}_model_confidence']
answers[pred_col][-1]['model_result'] = model_result
answers[pred_col] = sorted(answers[pred_col],
key=lambda x: x['confidence'],
reverse=True)
return answers
def explain(self):
answers = {}
for pred_col in self.evaluations:
answers[pred_col] = []
prediction_row = {
col: self.data[col][self.row_index]
for col in list(self.data.keys())
}
evaluation = self.evaluations[pred_col][self.row_index]
clusters = evaluation.explain(
self.transaction_output.transaction.lmd['column_stats']
[pred_col])
for cluster in clusters:
pct_confidence = round(cluster['confidence'] * 100)
predicted_value = cluster['value']
if f'{pred_col}_model_confidence' in prediction_row:
new_conf = round(
(prediction_row[f'{pred_col}_model_confidence'] * 3 +
cluster['confidence'] * 1) / 4, 4)
else:
new_conf = cluster['confidence']
if self.col_stats[pred_col]['data_type'] in (
DATA_TYPES.NUMERIC, DATA_TYPES.DATE):
value_range = [
cluster['buckets'][0], cluster['buckets'][-1]
]
range_pretty_start = value_range[0]
if range_pretty_start > 1000:
range_pretty_start = round(range_pretty_start)
if range_pretty_start > 100:
range_pretty_start = round(range_pretty_start, 2)
if range_pretty_start > 10:
range_pretty_start = round(range_pretty_start, 1)
elif range_pretty_start > pow(10, -3):
range_pretty_start = round(range_pretty_start, 6)
range_end_start = value_range[-1]
if range_end_start > 1000:
range_end_start = round(range_end_start)
if range_end_start > 100:
range_end_start = round(range_end_start, 2)
if range_end_start > 10:
range_end_start = round(range_end_start, 1)
elif range_end_start > pow(10, -3):
range_end_start = round(range_end_start, 6)
explanation = explain_prediction(
self.transaction_output.transaction.lmd,
prediction_row, cluster['confidence'], pred_col)
answers[pred_col].append({
'value':
predicted_value,
'range':
value_range,
'confidence':
new_conf,
'explanation':
explanation,
'explaination':
explanation,
'simple':
f'We are {pct_confidence}% confident the value of "{pred_col}" lies between {range_pretty_start} and {range_end_start}'
})
else:
explanation = explain_prediction(
self.transaction_output.transaction.lmd,
prediction_row, cluster['confidence'], pred_col)
answers[pred_col].append({
'value':
predicted_value,
'confidence':
new_conf,
'explanation':
explanation,
'explaination':
explanation,
'simple':
f'We are {pct_confidence}% confident the value of "{pred_col}" is {predicted_value}'
})
if self.transaction_output.input_confidence is not None:
for i in range(len(answers[pred_col])):
answers[pred_col][i]['confidence_influence_scores'] = {
'confidence_variation_score': [],
'column_names': []
}
for c in self.transaction_output.input_confidence:
answers[pred_col][i][
'confidence_influence_scores'][
'confidence_variation_score'].append(
self.transaction_output.
input_confidence[c])
answers[pred_col][i][
'confidence_influence_scores'][
'column_names'].append(str(c))
model_result = {'value': prediction_row[f'model_{pred_col}']}
if f'{pred_col}_model_confidence' in prediction_row:
model_result['confidence'] = prediction_row[
f'{pred_col}_model_confidence']
answers[pred_col][-1]['model_result'] = model_result
answers[pred_col] = sorted(answers[pred_col],
key=lambda x: x['confidence'],
reverse=True)
return answers