relationalai 0.13.0.dev0__py3-none-any.whl → 0.13.2__py3-none-any.whl

relationalai/semantics/lqp/primitives.py

@@ -0,0 +1,252 @@
+from relationalai.semantics.metamodel.types import digits_to_bits
+from relationalai.semantics.lqp import ir as lqp
+from relationalai.semantics.lqp.types import is_numeric
+from relationalai.semantics.lqp.utils import TranslationCtx, gen_unique_var, lqp_hash
+from relationalai.semantics.lqp.constructors import mk_primitive, mk_specialized_value, mk_type, mk_value
+
+rel_to_lqp = {
+    "=": "rel_primitive_eq",
+    "!=": "rel_primitive_neq",
+    "%": "rel_primitive_remainder",
+    "abs": "rel_primitive_abs",
+    "ceil": "rel_primitive_round_up",
+    "floor": "rel_primitive_round_down",
+    "pow": "rel_primitive_power",
+    "cbrt": "rel_primitive_cbrt",
+    "factorial": "rel_primitive_factorial",
+    "cos": "rel_primitive_cos",
+    "cosh": "rel_primitive_cosh",
+    "acos": "rel_primitive_acos",
+    "acosh": "rel_primitive_acosh",
+    "sin": "rel_primitive_sin",
+    "sinh": "rel_primitive_sinh",
+    "asin": "rel_primitive_asin",
+    "asinh": "rel_primitive_asinh",
+    "tan": "rel_primitive_tan",
+    "tanh": "rel_primitive_tanh",
+    "atan": "rel_primitive_atan",
+    "atanh": "rel_primitive_atanh",
+    "cot": "rel_primitive_cot",
+    "acot": "rel_primitive_acot",
+    "construct_date": "rel_primitive_construct_date",
+    "construct_date_from_datetime": "rel_primitive_datetime_date_convert",
+    "construct_datetime_ms_tz": "rel_primitive_construct_datetime",
+    "hash": "rel_primitive_hash_tuple_uint128",
+    "uuid_to_string": "rel_primitive_uuid_string",
+    "parse_uuid": "rel_primitive_parse_uuid",
+    "parse_date": "rel_primitive_parse_date",
+    "parse_datetime": "rel_primitive_parse_datetime",
+    "parse_decimal": "rel_primitive_parse_decimal",
+    "parse_int64": "rel_primitive_parse_int",
+    "parse_int128": "rel_primitive_parse_int128",
+    "parse_float": "rel_primitive_parse_float",
+    "string": "rel_primitive_string",
+    "starts_with": "rel_primitive_starts_with",
+    "ends_with": "rel_primitive_ends_with",
+    "contains": "rel_primitive_contains",
+    "num_chars": "rel_primitive_num_chars",
+    "substring": "rel_primitive_substring",
+    "like_match": "rel_primitive_like_match",
+    "lower": "rel_primitive_lowercase",
+    "upper": "rel_primitive_uppercase",
+    "concat": "rel_primitive_concat",
+    "replace": "rel_primitive_replace",
+    "strip": "rel_primitive_trim",
+    "date_year": "rel_primitive_date_year",
+    "date_quarter": "rel_primitive_date_quarterofyear",
+    "date_month": "rel_primitive_date_month",
+    "date_week": "rel_primitive_date_week",
+    "date_day": "rel_primitive_date_day",
+    "date_dayofyear": "rel_primitive_date_dayofyear",
+    "date_weekday": "rel_primitive_date_dayofweek",
+    "date_add": "rel_primitive_typed_add_date_period",
+    "date_subtract": "rel_primitive_typed_subtract_date_period",
+    "dates_period_days": "rel_primitive_date_days_between",
+    "datetime_now": "__pyrel_lqp_intrinsic_datetime_now",
+    "datetime_add": "rel_primitive_typed_add_datetime_period",
+    "datetime_subtract": "rel_primitive_typed_subtract_datetime_period",
+    "datetime_year": "rel_primitive_datetime_year",
+    "datetime_quarter": "rel_primitive_datetime_quarterofyear",
+    "datetime_month": "rel_primitive_datetime_month",
+    "datetime_week": "rel_primitive_datetime_week",
+    "datetime_day": "rel_primitive_datetime_day",
+    "datetime_dayofyear": "rel_primitive_datetime_dayofyear",
+    "datetime_hour": "rel_primitive_datetime_hour",
+    "datetime_minute": "rel_primitive_datetime_minute",
+    "datetime_second": "rel_primitive_datetime_second",
+    "datetime_weekday": "rel_primitive_datetime_dayofweek",
+    "datetimes_period_milliseconds": "rel_primitive_datetime_milliseconds_between",
+    "date_format": "rel_primitive_format_date",
+    "datetime_format": "rel_primitive_format_datetime",
+    "range": "rel_primitive_range",
+    "natural_log": "rel_primitive_natural_log",
+    "log": "rel_primitive_log",
+    "log2": "rel_primitive_log2",
+    "log10": "rel_primitive_log10",
+    "sqrt": "rel_primitive_sqrt",
+    "isinf": "rel_primitive_isinf",
+    "isnan": "rel_primitive_isnan",
+    "exp": "rel_primitive_natural_exp",
+    "erf": "rel_primitive_error_function",
+    "erfinv": "rel_primitive_error_function_inverse",
+    # Division is monotype, but only on the input args. Until we distinguish between input
+    # and output args, we can't use the same assertions for monotype-ness as the other ops.
+    "/": "rel_primitive_divide_monotype",
+    "levenshtein": "rel_primitive_levenshtein",
+    "split": "rel_primitive_string_split",
+    "split_part": "rel_primitive_string_split",
+    "regex_match": "rel_primitive_regex_match",
+    "infomap": "rel_primitive_infomap",
+    "louvain": "rel_primitive_louvain",
+    "label_propagation": "rel_primitive_async_label_propagation",
+}
+
+primitive_type_reorderings = {
+    "rel_primitive_like_match": [1, 0],
+}
+
+agg_to_lqp = {
+    "min": "rel_primitive_min",
+    "max": "rel_primitive_max",
+    "sum": "rel_primitive_add_monotype",
+    "count": "rel_primitive_add_monotype", # count is a sum of 1s
+    "rel_primitive_solverlib_ho_appl": "rel_primitive_solverlib_ho_appl",
+}
+
+rel_to_lqp_monotype = {
+    "+": "rel_primitive_add_monotype",
+    "-": "rel_primitive_subtract_monotype",
+    "*": "rel_primitive_multiply_monotype",
+    "<=": "rel_primitive_lt_eq_monotype",
+    ">=": "rel_primitive_gt_eq_monotype",
+    ">": "rel_primitive_gt_monotype",
+    "<": "rel_primitive_lt_monotype",
+    "//": "rel_primitive_trunc_divide_monotype",
+    "maximum": "rel_primitive_max",
+    "minimum": "rel_primitive_min",
+}
+
+# Insert extra terms where a raicode primitive expects more terms, and there are possible
+# defaults.
+def _extend_primitive_terms(name: str, terms: list[lqp.RelTerm], term_types: list[lqp.Type]) -> tuple[list[lqp.RelTerm], list[lqp.Type]]:
+    if name == "rel_primitive_parse_decimal" and len(terms) == 2:
+        assert term_types
+        py_precision = term_types[1].parameters[0].value
+        bit_value = mk_value(digits_to_bits(py_precision))
+        precision_value = mk_value(term_types[1].parameters[1].value)
+        terms = [
+            mk_specialized_value(bit_value),
+            mk_specialized_value(precision_value),
+            terms[0],
+            terms[1],
+        ]
+        term_types = [mk_type(lqp.TypeName.INT), mk_type(lqp.TypeName.INT), *term_types]
+
+    return (terms, term_types)
+
+# Reorder terms where the raicode primitive expects them in a different order.
+def _reorder_primitive_terms(name: str, terms: list[lqp.RelTerm], term_types: list[lqp.Type]) -> tuple[list[lqp.RelTerm], list[lqp.Type]]:
+    reordering = primitive_type_reorderings.get(name, None)
+    if reordering:
+        assert len(terms) == len(reordering), \
+            f"Primitive {name} expected {len(reordering)} terms, got {len(terms)}"
+        return ([terms[i] for i in reordering], [term_types[i] for i in reordering])
+
+    return (terms, term_types)
+
+# Check that the primitive terms have the expected types.
+# TODO: Fill out for all primitives, or rely on server-side type checking?
+def _assert_primitive_terms(name: str, terms: list[lqp.RelTerm], term_types: list[lqp.Type]):
+    if is_monotype(name):
+        # Make sure that the input terms have the same types
+        assert term_types.count(term_types[0]) == len(term_types), \
+            f"Expected all terms to have the same type for monotype operator " \
+            f"`{name}` but got {term_types} for terms {terms}"
+    elif name == "=":
+        # Allowed types are monotype, or both number types
+        monotype = term_types[0] == term_types[1]
+        numeric = is_numeric(term_types[0]) and is_numeric(term_types[1])
+        assert monotype or numeric, \
+            f"Expected types of `=`to be comparable but got `{[t.type_name for t in term_types]}`"
+
+def build_primitive(
+    name: str,
+    terms: list[lqp.RelTerm],
+    term_types: list[lqp.Type]
+) -> lqp.Formula:
+    lqp_name = relname_to_lqp_name(name)
+    terms, term_types = _extend_primitive_terms(lqp_name, terms, term_types)
+    terms, term_types = _reorder_primitive_terms(lqp_name, terms, term_types)
+    _assert_primitive_terms(lqp_name, terms, term_types)
+
+    # Handle intrinsics. To callers of `build_primitive` the distinction between intrinsic
+    # and primitive doesn't matter, so we don't want to burden them with that detail.
+    # Intrinsics are built-in definitions added by the LQP emitter, that user logic can just
+    # refer to.
+    if lqp_name == "__pyrel_lqp_intrinsic_datetime_now":
+        id = lqp.RelationId(id=lqp_hash(lqp_name), meta=None)
+        assert len(terms) == 1
+        assert isinstance(terms[0], lqp.Term)
+        return lqp.Atom(name=id, terms=[terms[0]], meta=None)
+
+    return mk_primitive(lqp_name, terms)
+
+def relname_to_lqp_name(name: str) -> str:
+    if name in rel_to_lqp:
+        return rel_to_lqp[name]
+    elif name in rel_to_lqp_monotype:
+        return rel_to_lqp_monotype[name]
+    else:
+        # If we don't have a mapping for the built-in, we just pass it through as-is.
+        return name
+
+def is_monotype(name: str) -> bool:
+    return name in rel_to_lqp_monotype
+
+# We take the name and type of the variable that we're summing over, so that we can generate
+# recognizable names for the variables in the reduce operation and preserve the type.
+def lqp_avg_op(ctx: TranslationCtx, op_name: str, sum_name: str, sum_type: lqp.Type) -> lqp.Abstraction:
+    count_type = mk_type(lqp.TypeName.INT)
+    vars = [
+        (gen_unique_var(ctx, sum_name), sum_type),
+        (gen_unique_var(ctx, "counter"), count_type),
+        (gen_unique_var(ctx, sum_name), sum_type),
+        (gen_unique_var(ctx, "one"), count_type),
+        (gen_unique_var(ctx, "sum"), sum_type),
+        (gen_unique_var(ctx, "count"), count_type),
+    ]
+
+    x1 = vars[0][0]
+    x2 = vars[1][0]
+    y1 = vars[2][0]
+    y2 = vars[3][0]
+    sumv = vars[4][0]
+    count = vars[5][0]
+
+    body = lqp.Conjunction(
+        args=[
+            mk_primitive("rel_primitive_add_monotype", [x1, y1, sumv]),
+            mk_primitive("rel_primitive_add_monotype", [x2, y2, count])
+        ],
+        meta=None
+    )
+    return lqp.Abstraction(vars=vars, value=body, meta=None)
+
+# Default handler for aggregation operations in LQP.
+def lqp_agg_op(ctx: TranslationCtx, op_name: str, aggr_arg_name: str, aggr_arg_type: lqp.Type) -> lqp.Abstraction:
+    x = gen_unique_var(ctx, f"x_{aggr_arg_name}")
+    y = gen_unique_var(ctx, f"y_{aggr_arg_name}")
+    z = gen_unique_var(ctx, f"z_{aggr_arg_name}")
+    ts = [(x, aggr_arg_type), (y, aggr_arg_type), (z, aggr_arg_type)]
+
+    name = agg_to_lqp.get(op_name, op_name)
+    body = mk_primitive(name, [x, y, z])
+
+    return lqp.Abstraction(vars=ts, value=body, meta=None)
+
+def lqp_operator(ctx: TranslationCtx, op_name: str, aggr_arg_name: str, aggr_arg_type: lqp.Type) -> lqp.Abstraction:
+    # TODO: Can we just pass through unknown operations?
+    if op_name not in agg_to_lqp:
+        raise NotImplementedError(f"Unsupported aggregation: {op_name}")
+
+    return lqp_agg_op(ctx, op_name, aggr_arg_name, aggr_arg_type)

relationalai/semantics/lqp/result_helpers.py

@@ -0,0 +1,202 @@
+from __future__ import annotations
+
+from collections import defaultdict
+from functools import reduce
+import re
+from typing import Any, Dict, List, Tuple, cast
+
+from pandas import DataFrame
+import pandas as pd
+
+from relationalai import debugging
+from relationalai.clients.result_helpers import format_columns, format_value, merge_columns, convert_hash_value, \
+    sort_data_frame_result
+from relationalai.tools.constants import Generation
+
+
+def format_results(results, result_cols:List[str]|None = None)  -> Tuple[DataFrame, List[Any]]:
+    with debugging.span("format_results"):
+        data_frame = DataFrame()
+        problems = defaultdict(
+            lambda: {
+                "message": "",
+                "path": "",
+                "start_line": None,
+                "end_line": None,
+                "report": "",
+                "code": "",
+                "severity": "",
+                "decl_id": "",
+                "name": "",
+                "output": "",
+                "end_character": None,
+                "start_character": None,
+                "props": {},
+            }
+        )
+
+        # Check if there are any results to process
+        if len(results.results):
+            ret_cols = result_cols or []
+            has_cols:List[DataFrame] = [DataFrame() for _ in range(0, len(ret_cols))]
+            key_len = 0
+
+            for result in results.results:
+                relation_id = result["relationId"]
+                result_frame = result["table"].to_pandas()
+                types = [
+                    t
+                    for t in result["relationId"].split("/")
+                    if t != "" and not t.startswith(":")
+                ]
+
+                # Process diagnostics
+                if "/:rel/:catalog/:diagnostic/" in relation_id:
+                    # Handle different types of diagnostics based on relation_id
+                    if "/:message/" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[row.iloc[0]]["message"] = row.iloc[1]
+                    elif "/:report/" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[row.iloc[0]]["report"] = row.iloc[1]
+                    elif "/:start/:line" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 2:
+                                problems[row.iloc[0]]["start_line"] = row.iloc[2]
+                    elif "/:end/:line" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 2:
+                                problems[row.iloc[0]]["end_line"] = row.iloc[2]
+                    elif "/:model" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[row.iloc[0]]["path"] = row.iloc[1]
+                    elif "/:severity" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[row.iloc[0]]["severity"] = row.iloc[1]
+                    elif "/:code" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[row.iloc[0]]["code"] = row.iloc[1]
+
+                # Process integrity constraint violations
+                elif "/:rel/:catalog/:ic_violation" in relation_id:
+                    if "/:decl_id" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])]["decl_id"] = (
+                                    row.iloc[1]
+                                )
+                                problems[convert_hash_value(row.iloc[0])]["code"] = (
+                                    "IC_VIOLATION"
+                                )
+                    elif "/:model" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])]["path"] = (
+                                    row.iloc[1]
+                                )
+                    elif "/:name" in relation_id:
+                        # Get the last segment of the relation_id as the name
+                        segments = [
+                            segment[1:]
+                            for segment in relation_id.split("/")[4:]
+                            if segment.startswith(":")
+                        ]
+                        for _, row in result_frame.iterrows():
+                            problems[convert_hash_value(row.iloc[0])]["name"] = segments[-1]
+                    elif "/:output" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])]["message"] = (
+                                    row.iloc[1]
+                                )
+                    elif "/:range/:end/:character" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])][
+                                    "end_character"
+                                ] = row.iloc[1]
+                    elif "/:range/:end/:line" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])]["end_line"] = (
+                                    row.iloc[1]
+                                )
+                    elif "/:range/:start/:character" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])][
+                                    "start_character"
+                                ] = row.iloc[1]
+                    elif "/:range/:start/:line" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])]["start_line"] = (
+                                    row.iloc[1]
+                                )
+                    elif "/:report" in relation_id:
+                        for _, row in result_frame.iterrows():
+                            if len(row) > 1:
+                                problems[convert_hash_value(row.iloc[0])]["report"] = (
+                                    row.iloc[1]
+                                )
+
+                elif "/:pyrel_error_attrs" in relation_id:
+                    assert len(result_frame.columns) == 3, \
+                        f"Error attribute with {len(result_frame.columns)} cols, expected 3"
+                    for _, row in result_frame.iterrows():
+                        id = convert_hash_value(row.iloc[0])
+                        problems[id]["code"] = "PYREL_ERROR"
+                        if row.iloc[1] == "message":
+                            problems[id]["message"] = row.iloc[2]
+                        elif row.iloc[1] == "severity":
+                            problems[id]["severity"] = row.iloc[2]
+                        else:
+                            props = cast(Dict, problems[id]["props"])
+                            props[row.iloc[1]] = format_value(row.iloc[2], relation_id.split("/")[-1])
+
+                elif "/:__pyrel_debug_watch" in relation_id:
+                    result_frame = format_columns(result_frame, types, Generation.QB)
+                    from relationalai.experimental.inspect import _print_watch_frame
+                    _print_watch_frame(result_frame)
+
+                # Process other results
+                else:
+                    result_frame = format_columns(result_frame, types, Generation.QB)
+                    result["table"] = result_frame
+                    if "/:output" in result["relationId"] \
+                            and "_cols_col" in result["relationId"]:
+                        # Match rows with an id like "/:output.*_cols_col[0-9]+"
+                        matched = re.search(r"_cols_col([0-9]+)", result["relationId"])
+                        assert matched, f"Column id not found for: {result['relationId']}"
+                        col_ix = int(matched.group(1))
+
+                        key_cols = [f"id{i}" for i in range(0, len(result_frame.columns) - 1)]
+                        key_len = len(key_cols)
+                        result_frame.columns = [*key_cols, f"v{col_ix}"]
+
+                        if has_cols[col_ix].empty:
+                            has_cols[col_ix] = result_frame
+                        else:
+                            has_cols[col_ix] = pd.concat([has_cols[col_ix], result_frame], ignore_index=True)
+                    elif ":output" in result["relationId"]:
+                        data_frame = pd.concat(
+                            [data_frame, result_frame], ignore_index=True
+                        )
+
+            if any(not col.empty for col in has_cols):
+                key_cols = [f"id{i}" for i in range(0, key_len)]
+                df_wide_reset = reduce(lambda left, right: merge_columns(left, right, key_cols), has_cols)
+                data_frame = df_wide_reset.drop(columns=key_cols)
+
+            data_frame = sort_data_frame_result(data_frame)
+
+            if len(ret_cols) and len(data_frame.columns) == len(ret_cols):
+                if result_cols is not None:
+                    data_frame.columns = result_cols[: len(data_frame.columns)]
+
+        return (data_frame, list(problems.values()))

relationalai/semantics/lqp/rewrite/annotate_constraints.py

@@ -0,0 +1,57 @@
+from __future__ import annotations
+
+from relationalai.semantics.metamodel import builtins
+from relationalai.semantics.metamodel.ir import Node, Model, Require
+from relationalai.semantics.metamodel.compiler import Pass
+from relationalai.semantics.metamodel.rewrite.discharge_constraints import (
+    DischargeConstraintsVisitor
+)
+from relationalai.semantics.lqp.rewrite.functional_dependencies import (
+    is_valid_unique_constraint, normalized_fd
+)
+
+_DISABLE_CONSTRAINT_DECLARATIONS = True
+
+class AnnotateConstraints(Pass):
+    """
+    Extends `DischargeConstraints` pass by discharging only those Require nodes that cannot
+    be declared as constraints in LQP.
+
+    More precisely, the pass annotates Require nodes depending on how they should be
+    treated when generating code:
+     * `@declare_constraint` if the Require represents a constraint that can be declared in LQP.
+     * `@discharge` if the Require represents a constraint that should be dismissed during
+       code generation. Namely, when it cannot be declared in LQP and uses one of the
+       `unique`, `exclusive`, `anyof` builtins. These nodes are removed from the IR model
+       in the Flatten pass.
+    """
+
+    def rewrite(self, model: Model, options: dict = {}) -> Model:
+        return AnnotateConstraintsRewriter().walk(model)
+
+
+class AnnotateConstraintsRewriter(DischargeConstraintsVisitor):
+    """
+    Visitor marks all nodes which should be removed from IR model with `discharge` annotation.
+    """
+
+    def _should_be_declarable_constraint(self, node: Require) -> bool:
+        if _DISABLE_CONSTRAINT_DECLARATIONS:
+            return False
+        if not is_valid_unique_constraint(node):
+            return False
+        # Currently, we only declare non-structural functional dependencies.
+        fd = normalized_fd(node)
+        assert fd is not None  # already checked by _is_valid_unique_constraint
+        return not fd.is_structural
+
+    def handle_require(self, node: Require, parent: Node):
+        if self._should_be_declarable_constraint(node):
+            return node.reconstruct(
+                node.engine,
+                node.domain,
+                node.checks,
+                node.annotations | [builtins.declare_constraint_annotation]
+            )
+
+        return super().handle_require(node, parent)