Built-in Functions¶

AISL includes over 40 built-in functions that are always available without declaration. They cover machine learning metrics, statistics, data operations, and LLM interaction.

Categories¶

Category	Description
ML Metrics	Classification and evaluation metrics
Statistics	Statistical aggregation and tests
Information Theory	Entropy, divergence, and mutual information
Consensus	Voting, agreement, and frequency analysis
Data Operations	Extracting, filtering, and transforming data
Response Parsing	Parsing and extracting values from text
LLM Queries	Querying language models

Quick reference¶

All built-in functions are called using standard function call syntax:

let result = function_name(arg1, arg2, ...);

ML Metrics¶

Function	Description
`accuracy(y_true, y_pred)`	Classification accuracy
`f1_score(y_true, y_pred)`	F1 score
`precision(y_true, y_pred)`	Precision
`recall(y_true, y_pred)`	Recall
`confusion_matrix(y_true, y_pred)`	Confusion matrix
`roc_curve(y_true, y_pred)`	ROC curve data
`brier_score_model(y_true, y_pred)`	Brier score (whole set)
`brier_score_element(y_true, y_pred)`	Per-element Brier scores

Statistics¶

Function	Description
`mean(list)`	Arithmetic mean
`median(list)`	Median value
`variance(list)`	Sample variance
`standard_deviation(list)`	Sample standard deviation
`percentile(list, p)`	p-th percentile
`confidence_interval(list, level)`	Confidence interval bounds
`paired_t_test(mean_diff, std, n)`	Paired t-statistic
`t_test_p_value(t_stat, n)`	P-value from t-statistic

Information Theory¶

Function	Description
`binary_entropy(p)`	Shannon entropy for binary probability
`mean_entropy(probabilities)`	Average entropy
`aggregate_entropy(dataset)`	Per-model, per-question entropy
`kl_divergence(questions, answers, aggregate)`	KL divergence
`symmetric_kl_divergence(answers, aggregate)`	Symmetric KL divergence
`mutual_information_heatmap(data)`	Mutual information between models

Consensus¶

Function	Description
`majority_vote(list)`	Most common element
`consensus_threshold(list, threshold)`	Consensus check
`mode(list)`	Statistical mode
`agreement_rate(list)`	Rate of the most common element
`unique_values(list)`	Distinct values
`value_counts(list)`	Frequency of each value

Data Operations¶

Function	Description
`get(data, field)`	Extract a single field
`get_all(data, field)`	Extract a field from all rows
`filter_by_key(data, key, value)`	Filter rows by field value
`flatten(array)`	Flatten a multi-dimensional array
`argmax(array)`	Index of maximum value per element
`convert_to_one_hot(array)`	Convert to one-hot encoding
`fill(value, length)`	Create array of repeated values
`sum(list)`	Sum of elements
`size(list)`	Number of elements
`histogram(data, question, answer)`	Frequency distribution
`box_plot(data, question, answer, score)`	Box plot data

Response Parsing¶

Function	Description
`parse_json(text)`	Parse a JSON string
`extract_key(text, key, default)`	Extract a key from text
`extract_number(text)`	Extract the first number
`extract_sentiment(text)`	Classify sentiment

LLM Queries¶

Function	Description
`query_model(model, prompt, context)`	Query a declared model
`llm_query(prompt, context)`	Query the default LLM