def _generate_example(
component_name: str
) -> Tuple[List[pd.DataFrame], pd.DataFrame, str, Graph]:
while True:
try:
inputs, replay_map = datagen_dict[component_name]()
# Abstractions for the individual inputs.
g_inputs = [DataFrameGraph(i) for i in inputs]
strategy = RandomizedGraphStrategy()
gen = generator_dict[component_name]
output, program, graph, output_graph = gen.with_env(
strategy=strategy, replay=replay_map).call(*inputs,
*g_inputs,
datagen=True)
if 0 in output.shape:
raise AssertionError("Got empty dataframe")
# Populate the placeholders
program = program.format(
**{f"inp{i}": f"inp{i}"
for i in range(1,
len(inputs) + 1)})
return inputs, output, program, graph
except Exception as e:
pass
def test_df_graph(self):
from gauss.domains.rlang.graphs import DataFrameGraph
df = pd.DataFrame([['a', 'b', 'e'], ['c', 'd', 'f']], columns=['C1', 'C2', 'C3'])
df_graph = DataFrameGraph(df)
# Check if all the nodes have been created.
self.assertListEqual(list(df.columns), [c.value for c in df_graph.columns])
self.assertListEqual(list(df.index), [c.value for c in df_graph.index])
for row_df, row_df_graph in zip(df.values, df_graph.values):
self.assertListEqual(list(row_df), [v.value for v in row_df_graph])
def test_gather(self):
from gauss.domains.rlang.generators import gen_gather, DataFrameGraph, RInterpreter
gather = RInterpreter.gather
df = pd.DataFrame([['a', 'b', 'e'], ['c', 'd', 'f']], columns=['C1', 'C2', 'C3'])
result, call_str, graph, res_graph = gen_gather.call(df, DataFrameGraph(df))
# Result should not be equal to the input
self.assertRaises(AssertionError, pd.testing.assert_frame_equal, df, result)
# Call str should evaluate to the result
pd.testing.assert_frame_equal(result, eval(call_str.format(inp1='df')))
def test_groupby(self):
from gauss.domains.rlang.generators import gen_group_by_summarise, DataFrameGraph, RInterpreter
group_by = RInterpreter.group_by
summarise = RInterpreter.summarise
df = pd.DataFrame([['A', 100], ['A', 200], ['B', 300]], columns=['C1', 'C2'])
g_df = DataFrameGraph(df)
result, call_str, graph, res_graph = gen_group_by_summarise.with_env(ignore_exceptions=False).call(df, g_df)
# Result should not be equal to the input
self.assertRaises(AssertionError, pd.testing.assert_frame_equal, df, result)
# Call str should evaluate to the result
pd.testing.assert_frame_equal(result, eval(call_str.format(inp1='df')))
def test_inner_join(self):
from gauss.domains.rlang.generators import gen_inner_join, DataFrameGraph, RInterpreter
inner_join = RInterpreter.inner_join
df1 = pd.DataFrame([['a', 'b', 'c'], ['d', 'g', 'c'], ['f', 'b', 'h']], columns=['c1', 'c2', 'c3'])
df2 = pd.DataFrame([['x', 'g', 'z'], ['w', 'b', 'u'], ['y', 'g', 'j']], columns=['c4', 'c2', 'c5'])
result, call_str, graph, res_graph = gen_inner_join.call(df1, df2, DataFrameGraph(df1), DataFrameGraph(df2))
# Result should not be equal to the input
self.assertRaises(AssertionError, pd.testing.assert_frame_equal, df1, result)
self.assertRaises(AssertionError, pd.testing.assert_frame_equal, df2, result)
# Call str should evaluate to the result
pd.testing.assert_frame_equal(result, eval(call_str.format(inp1='df1', inp2='df2')))
def gen_inner_join(df1: pd.DataFrame,
df2: pd.DataFrame,
g_df1: DataFrameGraph,
g_df2: DataFrameGraph,
datagen: bool = False):
"""
INNER_JOIN
------
Example:
inner_join(df1, df2)
c1 c2 c3 c4 c2 c5 c1 c2 c3 c4 c5
0 a b c 0 x g z 0 a b c w u
1 d g c 1 w b u -> 1 f b h w u
2 f b h , 2 y g j 2 d g c x z
3 d g c y j
---------------
Graph Abstraction:
- EQUAL edges from all the columns in df1 and df2 to the corresponding column in the output.
- EQUAL edges from all the rows in df1 and df2 to the corresponding row in the output, if included.
- EQUAL edge between the input deletion nodes and the output deletion node.
- DELETE edges from all the non-included cells in both df1 and df2 to the deletion node of the output.
"""
result = RInterpreter.inner_join(df1, df2)
call_str = f"inner_join({{inp1}}, {{inp2}})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df1, g_df2, g_res])
added_edges: List[Edge] = []
col_map_df1 = {c.value: c
for c in g_df1.columns
} # Map from df1's columns to their column nodes
col_map_df2 = {c.value: c
for c in g_df2.columns
} # Map from df2's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edges from all the columns in df1 and df2 to the corresponding column in the output.
for c, node in itertools.chain(col_map_df1.items(), col_map_df2.items()):
added_edges.append(Edge(node, col_map_res[c], ELabel.EQUAL))
# Get the merge cols
merge_cols = list(set(col_map_df1.keys()) & set(col_map_df2.keys()))
# Get the indices for each value tuple for df1 and df2 and result
value_dict_df1 = collections.defaultdict(list)
value_dict_df2 = collections.defaultdict(list)
value_dict_res = collections.defaultdict(list)
for idx, values in zip(df1.index, df1.loc[:, merge_cols].values):
values = tuple(values)
value_dict_df1[values].append(idx)
for idx, values in zip(df2.index, df2.loc[:, merge_cols].values):
values = tuple(values)
value_dict_df2[values].append(idx)
for idx, values in zip(result.index, result.loc[:, merge_cols].values):
values = tuple(values)
value_dict_res[values].append(idx)
# - EQUAL edges from all the rows in df1 and df2 to the corresponding row in the output, if included.
# - DELETE edges from all the non-included cells in both df1 and df2 to the deletion node of the output.
deleted_df1 = set(df1.index)
deleted_df2 = set(df2.index)
for value, res_indices in value_dict_res.items():
df1_indices = value_dict_df1[value]
df2_indices = value_dict_df2[value]
deleted_df1.difference_update(df1_indices)
deleted_df2.difference_update(df2_indices)
for idx_res, (idx1,
idx2) in zip(res_indices,
itertools.product(df1_indices,
df2_indices)):
for c in df1.columns:
added_edges.append(
Edge(g_df1.loc[idx1, c], g_res.loc[idx_res, c],
ELabel.EQUAL))
for c in df2.columns:
added_edges.append(
Edge(g_df2.loc[idx2, c], g_res.loc[idx_res, c],
ELabel.EQUAL))
for c in merge_cols:
added_edges.append(
Edge(g_df1.loc[idx1, c], g_df2.loc[idx2, c], ELabel.EQUAL))
added_edges.append(
Edge(g_df2.loc[idx2, c], g_df1.loc[idx1, c], ELabel.EQUAL))
for idx1 in deleted_df1:
for c in df1.columns:
added_edges.append(
Edge(g_df1.loc[idx1, c], g_res.deletion_node, ELabel.DELETE))
for idx2 in deleted_df2:
for c in df2.columns:
added_edges.append(
Edge(g_df2.loc[idx2, c], g_res.deletion_node, ELabel.DELETE))
# - EQUAL edge between the input deletion nodes and the output deletion node.
added_edges.append(
Edge(g_df1.deletion_node, g_res.deletion_node, ELabel.EQUAL))
added_edges.append(
Edge(g_df2.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
# No arguments for this component.
return result, call_str, graph, g_res
def gen_filter(df: pd.DataFrame, g_df: DataFrameGraph, datagen: bool = False):
"""
FILTER
------
Example:
filter(df, 'C1 > 3')
---------------
df result
C1 C2 C3 C1 C2 C3
0 3 a d --> 0 4 b e
1 4 b e 1 5 c f
2 5 c f
---------------
Graph Abstraction:
- EQUAL edges between the all columns and cells in the input that are preserved to the
corresponding nodes of the output.
- Additional dependency edges from the cells of the column used in the filtering condition (C1 in the example).
"""
cands_column = g_df.columns
mode = SelectConst(["equality-inequality", "relop"], uid="filter_mode")
if mode == "equality-inequality":
column = SelectNode(cands_column, uid="filter_column_eq")
all_values = set(df[column])
value = SelectConst(list(all_values), uid="filter_value_eq")
op = SelectConst(["==", "!="], uid="filter_eq_op")
filter_expr = f"{column} {op} {value!r}"
else:
numeric_cols = set(df.select_dtypes('number').columns)
column = SelectNode(
[c for c in cands_column if c.value in numeric_cols],
uid="filter_column_relop")
all_values = set(df[column])
value = SelectConst(list(all_values), uid="filter_value_relop")
op = SelectConst(["<", ">"], uid="filter_relop")
filter_expr = f"{column} {op} {value!r}"
result = RInterpreter.filter_(df, filter_expr, reset_index=False)
filtered_indices = list(result.index)
removed_indices = list(set(df.index) - set(filtered_indices))
result = result.reset_index(drop=True)
call_str = f"filter({{inp1}}, {filter_expr!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
if op == "==":
dep_label = ELabel.DEPENDENT_EQ
elif op == "!=":
dep_label = ELabel.DEPENDENT_INEQ
elif op == "<":
dep_label = ELabel.DEPENDENT_LT
else:
dep_label = ELabel.DEPENDENT_GT
deletion_node_out = g_res.deletion_node
# - EQUAL edges for all columns.
for c in col_map_df:
added_edges.append(Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
# - EQUAL edges for kept rows
for index, v1, v2 in zip(filtered_indices, g_df.loc[filtered_indices,
c], g_res.loc[:,
c]):
interm_node = graph.create_intermediate_node(v2.value)
added_edges.append(
Edge(g_df.loc[index, column], interm_node, dep_label))
added_edges.append(Edge(v1, interm_node, ELabel.EQUAL))
added_edges.append(Edge(interm_node, v2, ELabel.EQUAL))
# - Mark the rest as deleted
for v in g_df.loc[removed_indices, c]:
added_edges.append(Edge(v, deletion_node_out, ELabel.DELETE))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
if mode == 'equality-inequality':
for c_node in cands_column:
if column == c_node.value:
tagged_edges.append(
TaggedEdge(c_node, c_node, "SELECTED@filter_column_eq"))
else:
tagged_edges.append(
TaggedEdge(c_node, c_node,
"NOT_SELECTED@filter_column_eq"))
for cand_op in ["==", "!="]:
if cand_op == op:
graph.add_tags([f"SELECTED@{cand_op}@filter_eq_op"])
else:
graph.add_tags([f"NOT_SELECTED@{cand_op}@filter_eq_op"])
else:
for c_node in cands_column:
if column == c_node.value:
tagged_edges.append(
TaggedEdge(c_node, c_node, "SELECTED@filter_column_relop"))
else:
tagged_edges.append(
TaggedEdge(c_node, c_node,
"NOT_SELECTED@filter_column_relop"))
for cand_op in ["<", ">"]:
if cand_op == op:
graph.add_tags([f"SELECTED@{cand_op}@filter_relop"])
else:
graph.add_tags([f"NOT_SELECTED@{cand_op}@filter_relop"])
graph.add_tagged_edges(tagged_edges)
for cand_mode in ["equality-inequality", "relop"]:
if cand_mode == mode:
graph.add_tags([f"SELECTED@{cand_mode}@filter_mode"])
else:
graph.add_tags([f"NOT_SELECTED@{cand_mode}@filter_mode"])
return result, call_str, graph, g_res
def gen_mutate(df: pd.DataFrame, g_df: DataFrameGraph, datagen: bool = False):
"""
MUTATE
------
Example:
mutate(df, new_col_name='C4', operation='normalize', col_args='C1')
---------------
df result
C1 C2 C3 C1 C2 C3 C4
0 3 a d --> 0 3 a d 0.250000
1 4 b e 1 4 b e 0.333333
2 5 c f 2 5 c f 0.416666
mutate(df, new_col_name='C4', operation='div', col_args=['C1', 'C2'])
---------------
df result
C1 C2 C3 C1 C2 C3 C4
0 3 5 d --> 0 3 a d 0.6
1 4 10 e 1 4 b e 0.4
2 5 20 f 2 5 c f 0.25
---------------
Graph Abstraction:
- EQUAL edges between the all columns and cells in the input to the corresponding nodes of the output.
- If normalize, SUM and DIV edges representing the computation. A unique sum intermediate node is created for each
cell of the column being normalized.
"""
cands_cols = g_df.columns
operation = SelectConst(["normalize", "div"], uid="mutate_operation")
new_col_name = FreshColumn(uid="mutate_new_col_name")
if operation == "normalize":
col_arg = SelectNode(cands_cols, uid="mutate_col_args_normalize")
col_args = [col_arg]
result = RInterpreter.mutate(df,
new_col_name=new_col_name,
operation=operation,
col_args=col_arg)
call_str = f"mutate({{inp1}}, new_col_name={new_col_name!r}, operation={operation!r}, col_args={col_arg!r})"
else:
col_arg1, col_arg2 = OrderedSubsetNode(cands_cols,
uid="mutate_col_args_div",
min_len=2,
max_len=2)
col_args = [col_arg1, col_arg2]
result = RInterpreter.mutate(df,
new_col_name=new_col_name,
operation=operation,
col_args=col_args)
call_str = f"mutate({{inp1}}, new_col_name={new_col_name!r}, operation={operation!r}, col_args={col_args!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edges between the corresponding columns as all columns are preserved.
# - EQUAL edges between the cells that are preserved.
for c in col_map_df:
added_edges.append(Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
for v1, v2 in zip(g_df.loc[:, c], g_res.loc[:, c]):
added_edges.append(Edge(v1, v2, ELabel.EQUAL))
if operation == 'normalize':
summation = df[col_args[0]].sum()
for cell_node, res_node in zip(g_df.loc[:, col_args[0]],
g_res.loc[:, new_col_name]):
# Add the sum edges
interm_node_sum = graph.create_intermediate_node(summation)
for c in g_df.loc[:, col_args[0]]:
added_edges.append(Edge(c, interm_node_sum, ELabel.SUM))
interm_node_div = graph.create_intermediate_node(res_node.value)
added_edges.append(
Edge(interm_node_sum, interm_node_div, ELabel.DIV))
added_edges.append(Edge(cell_node, interm_node_div, ELabel.DIV))
added_edges.append(Edge(interm_node_div, res_node, ELabel.EQUAL))
else:
for cell_nodes, res_node in zip(g_df.loc[:, col_args].values,
g_res.loc[:, new_col_name]):
interm_node = graph.create_intermediate_node(res_node.value)
added_edges.append(Edge(interm_node, res_node, ELabel.EQUAL))
for n in cell_nodes:
added_edges.append(Edge(n, interm_node, ELabel.DIV))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
if operation == 'normalize':
for c_node in cands_cols:
if c_node.value in col_args:
tagged_edges.append(
TaggedEdge(c_node, c_node,
"SELECTED@mutate_col_args_normalize"))
else:
tagged_edges.append(
TaggedEdge(c_node, c_node,
"NOT_SELECTED@mutate_col_args_normalize"))
else:
for c_node in cands_cols:
if c_node.value in col_args:
tagged_edges.append(
TaggedEdge(c_node, c_node, "SELECTED@mutate_col_args_div"))
else:
tagged_edges.append(
TaggedEdge(c_node, c_node,
"NOT_SELECTED@mutate_col_args_div"))
graph.add_tags([
f"SELECTED@{cand}@mutate_operation"
if cand == operation else f"NOT_SELECTED@{cand}@mutate_operation"
for cand in ["normalize", "div"]
])
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def gen_select(df: pd.DataFrame, g_df: DataFrameGraph, datagen: bool = False):
"""
SELECT
------
Example:
select(df, columns_keep=None, columns_remove=['C3'])
---------------
df result
C1 C2 C3 C1 C2
0 3 a d --> 0 3 a
1 4 b e 1 4 b
2 5 c f 2 5 c
---------------
Graph Abstraction:
- EQUAL edges between the columns, cells of the preserved columns and the corresponding cells in the output.
- DELETE edges for the removed columns and their cells.
"""
cands_cols = list(g_df.columns)
choice = SelectConst([True, False], uid="select_keep_or_remove")
if choice:
columns_keep = list(
SubsetNode(cands_cols,
uid="select_columns_keep",
allow_empty=False))
columns_remove = None
else:
columns_keep = None
columns_remove = list(
SubsetNode(cands_cols,
uid="select_columns_remove",
allow_empty=False))
result = RInterpreter.select(df,
columns_keep=columns_keep,
columns_remove=columns_remove)
call_str = f"select({{inp1}}, columns_keep={columns_keep!r}, columns_remove={columns_remove!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
if choice:
kept_cols = set(columns_keep)
removed_cols = [c for c in df.columns if c not in kept_cols]
else:
removed_cols = set(columns_remove)
kept_cols = [c for c in df.columns if c not in removed_cols]
# - EQUAL edges between the corresponding preserved columns.
# - EQUAL edges between the cells that are preserved.
for c in kept_cols:
added_edges.append(Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
for v1, v2 in zip(g_df.loc[:, c], g_res.loc[:, c]):
added_edges.append(Edge(v1, v2, ELabel.EQUAL))
# - DELETE edges for the deleted columns and their cells.
for c in removed_cols:
added_edges.append(
Edge(col_map_df[c], g_res.deletion_node, ELabel.DELETE))
for v in g_df.loc[:, c]:
added_edges.append(Edge(v, g_res.deletion_node, ELabel.DELETE))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
if choice:
for c_node in cands_cols:
if c_node.value in columns_keep:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@select_columns_keep"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@select_columns_keep"))
else:
for c_node in cands_cols:
if c_node.value in columns_remove:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@select_columns_remove"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@select_columns_remove"))
for cand_choice in [True, False]:
if cand_choice == choice:
graph.add_tags([f"SELECTED@{cand_choice}@select_keep_or_remove"])
else:
graph.add_tags(
[f"NOT_SELECTED@{cand_choice}@select_keep_or_remove"])
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def gen_spread(df: pd.DataFrame, g_df: DataFrameGraph, datagen: bool = False):
"""
SPREAD
------
Example:
spread(columns='var', values='value')
---------------
df result
C1 var value C1 C2 C3
0 c C2 b 0 a d e
1 a C2 d --> 1 c b NaN
2 a C3 e
---------------
Graph Abstraction:
- EQUAL edge between the nodes in the `columns` column and the column node of the result.
- EQUAL edge between the `index` column and the corresponding column node of the result.
- EQUAL edge between the cells of `values` column and the corresponding cells in the result.
- EQUAL edge between the input deletion node and the output deletion node.
"""
cands_columns = g_df.columns
columns = SelectNode(cands_columns, uid="spread_columns")
cands_values = [c for c in g_df.columns if c.value != columns]
values = SelectNode(cands_values, uid="spread_values")
index = [c for c in df.columns if c != columns and c != values
] # The columns that will remain as is.
result = RInterpreter.spread(df, columns=columns, values=values)
call_str = f"spread({{inp1}}, columns={columns!r}, values={values!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edge between the nodes in the `columns` column and the column node of the result.
for cell in g_df.loc[:, columns]:
added_edges.append(Edge(cell, col_map_res[cell.value], ELabel.EQUAL))
# - EQUAL edge between the `index` column and the corresponding column node of the result.
# - EQUAL edge between the cells of `index` column and the corresponding cells in the result.
for c in index:
added_edges.append(Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
value_map = {n.value: n for n in g_res.loc[:, c]}
for cell in g_df.loc[:, c]:
added_edges.append(Edge(cell, value_map[cell.value], ELabel.EQUAL))
# - EQUAL edge between the cells of `values` column and the corresponding cells in the result.
for idx_vals, col_val, df_val_node in zip(
df[index].values if len(index) > 0 is not None else df.index,
df.loc[:, columns], g_df.loc[:, values]):
if len(index) == 0:
filtered = [g_res.loc[idx_vals, col_val]]
else:
idx_mask = True
for idx_val, idx in zip(idx_vals, index):
idx_mask = idx_mask & (result[idx] == idx_val)
filtered = list(g_res.loc[idx_mask][col_val])
assert len(filtered) == 1
res_val_node = filtered[0]
added_edges.append(Edge(df_val_node, res_val_node, ELabel.EQUAL))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
for c_node in cands_columns:
if c_node.value == columns:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@spread_columns"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@spread_columns"))
for c_node in cands_values:
if c_node.value == values:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@spread_values"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@spread_values"))
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def gen_separate(df: pd.DataFrame,
g_df: DataFrameGraph,
datagen: bool = False):
"""
SEPARATE
------
Example:
separate(split_col='C2', into=["C3", "C4"])
---------------
df result
C1 C2 C1 C3 C4
0 3 a_d 0 3 a d
1 4 b_e 1 4 b e
2 5 c_f 2 5 c f
---------------
Graph Abstraction:
- EQUAL edges between the columns, cells of the preserved columns and the corresponding cells in the output.
- STR_SPLIT edges between concerned cells.
"""
cands_cols = g_df.columns
split_col = SelectNode(cands_cols, min_len=2, uid="separate_split_col")
max_into_len = max([
len(re.compile("[^a-zA-Z0-9]+").split(str(x))) for x in df[split_col]
])
if max_into_len <= 1:
raise ExceptionAsContinue
into = [FreshColumn(uid="separate_into") for _ in range(max_into_len)]
result = RInterpreter.separate(df, split_col=split_col, into=into)
call_str = f"separate({{inp1}}, split_col={split_col!r}, into={into!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edges between columns that are unused and their cells to their counterparts.
unused_cols = set(col_map_df)
unused_cols.difference_update(split_col)
for c in col_map_df:
if c != split_col:
added_edges.append(
Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
for v1, v2 in zip(g_df.loc[:, c], g_res.loc[:, c]):
added_edges.append(Edge(v1, v2, ELabel.EQUAL))
for df_node, res_nodes in zip(g_df.loc[:, split_col],
g_res.loc[:, into].values):
for n in res_nodes:
interm_node = graph.create_intermediate_node(n.value)
added_edges.append(Edge(interm_node, n, ELabel.EQUAL))
added_edges.append(Edge(df_node, interm_node, ELabel.STR_SPLIT))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
for c_node in cands_cols:
if c_node.value == split_col:
tagged_edges.append(
TaggedEdge(c_node, c_node, "SELECTED@separate_split_col"))
else:
tagged_edges.append(
TaggedEdge(c_node, c_node, "NOT_SELECTED@separate_split_col"))
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def gen_unite(df: pd.DataFrame, g_df: DataFrameGraph, datagen: bool = False):
"""
UNITE
------
Example:
unite(cols=['C2', 'C3'], new_col_name='C4')
---------------
df result
C1 C2 C3 C1 C4
0 3 a d --> 0 3 a_d
1 4 b e 1 4 b_e
2 5 c f 2 5 c_f
---------------
Graph Abstraction:
- EQUAL edges between the columns, cells of the preserved columns and the corresponding cells in the output.
- STR_JOIN edges between concerned cells.
"""
cands_cols = g_df.columns
cols = list(OrderedSubsetNode(cands_cols, min_len=2, uid="unite_cols"))
new_col_name = FreshColumn(uid="unite_new_col_name")
result = RInterpreter.unite(df, cols=cols, new_col_name=new_col_name)
call_str = f"unite({{inp1}}, cols={cols!r}), new_col_name={new_col_name!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edges between columns that are unused and their cells to their counterparts.
unused_cols = set(col_map_df)
unused_cols.difference_update(cols)
for c in unused_cols:
added_edges.append(Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
for v1, v2 in zip(g_df.loc[:, c], g_res.loc[:, c]):
added_edges.append(Edge(v1, v2, ELabel.EQUAL))
for df_nodes, res_node in zip(g_df.loc[:, cols].values,
g_res.loc[:, new_col_name]):
interm_node = graph.create_intermediate_node(res_node.value)
for n in df_nodes:
added_edges.append(Edge(n, interm_node, ELabel.STR_JOIN))
added_edges.append(Edge(interm_node, res_node, ELabel.EQUAL))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
for c_node in cands_cols:
if c_node.value in cols:
tagged_edges.append(
TaggedEdge(c_node, c_node, "SELECTED@unite_cols"))
else:
tagged_edges.append(
TaggedEdge(c_node, c_node, "NOT_SELECTED@unite_cols"))
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def gen_gather(df: pd.DataFrame, g_df: DataFrameGraph, datagen: bool = False):
"""
GATHER
------
Example:
gather(id_vars=['C1'], value_vars=['C2', 'C3'], var_name='var', value_name='value')
df result
C1 C2 C3 C1 var value
0 a b e --> 0 a C2 b
1 c d f 1 c C2 d
2 a C3 e
3 c C3 f
---------------
Graph Abstraction:
- EQUAL edges between id_var columns and corresponding columns in output.
- EQUAL edges between value_var columns and corresponding cells in output.
- EQUAL edges between cells of id_var and value_var columns to the corresponding cells in output.
- EQUAL edge between the input deletion node and the output deletion node.
- DELETE edge between the columns not in id_vars and value_vars to the output deletion node.
- DELETE edge between the cells of columns not in id_vars and value_vars to the output deletion node.
"""
cands_id_vars = g_df.columns
id_vars = list(
SubsetNode(cands_id_vars, uid="gather_id_vars", allow_empty=True))
cands_value_vars = [c for c in g_df.columns if c.value not in id_vars]
value_vars = list(SubsetNode(cands_value_vars, uid="gather_value_vars"))
var_name = FreshColumn(uid="gather_var_name")
value_name = FreshColumn(uid="gather_value_name")
result = RInterpreter.gather(df,
id_vars=id_vars,
value_vars=value_vars,
var_name=var_name,
value_name=value_name)
call_str = f"gather({{inp1}}, id_vars={id_vars!r}, value_vars={value_vars!r}, " \
f"var_name={var_name!r}, value_name={value_name!r})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edges between id_var columns and corresponding columns in output.
for c1, c2 in ((col_map_df[c], col_map_res[c]) for c in id_vars or []):
added_edges.append(Edge(c1, c2, ELabel.EQUAL))
# - EQUAL edges between value_var columns and corresponding cells in output.
for var_node in g_res.loc[:, var_name]:
added_edges.append(
Edge(col_map_df[var_node.value], var_node, ELabel.EQUAL))
# - EQUAL edges between cells of id_var and value_var columns to the corresponding cells in output.
for col in id_vars or []:
df_nodes_concat = list(g_df.loc[:, col]) * len(value_vars)
for n1, n2 in zip(df_nodes_concat, g_res.loc[:, col]):
added_edges.append(Edge(n1, n2, ELabel.EQUAL))
value_nodes_concat = sum((list(g_df.loc[:, c]) for c in value_vars), [])
for n1, n2 in zip(value_nodes_concat, g_res.loc[:, value_name]):
added_edges.append(Edge(n1, n2, ELabel.EQUAL))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# - DELETE edge between the columns not in id_vars and value_vars to the output deletion node.
# - DELETE edge between the cells of columns not in id_vars and value_vars to the output deletion node.
unused_cols = [
c for c in df.columns if c not in id_vars and c not in value_vars
]
for col in unused_cols:
added_edges.append(
Edge(col_map_df[col], g_res.deletion_node, ELabel.DELETE))
for val_node in g_df.loc[:, col]:
added_edges.append(
Edge(val_node, g_res.deletion_node, ELabel.DELETE))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
for c_node in cands_id_vars:
if id_vars is not None and c_node.value in id_vars:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@gather_id_vars"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@gather_id_vars"))
for c_node in cands_value_vars:
if c_node.value in value_vars:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@gather_value_vars"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@gather_value_vars"))
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def gen_group_by_summarise(df: pd.DataFrame,
g_df: DataFrameGraph,
datagen: bool = False):
"""
GROUP_BY (+ SUMMARISE)
------
Example:
group_by(group_cols=['C1']).summarise(summaries={"C3": ("C2", "mean")})
df result
C1 C2 C1 C3
0 A 100 0 A 150
1 A 200 --> 1 B 300
2 B 300
---------------
Graph Abstraction:
- EQUAL edges between group columns and corresponding columns in output.
- EQUAL edges between the cells of the group columns and the corresponding cells in the output.
- SUM/MEAN/COUNT edges between the cells of aggregated columns (or group columns in case of 'count') and the
resulting cells in the output.
- EQUAL edge between the input deletion node and the output deletion node.
- DELETE edges between the non-group cols and non-aggregated columns and their cells
to the deletion node of the output.
- DELETE edge between the column of the aggregated column (in case of sum/mean) and the deletion node of the output.
"""
cands_group_cols = g_df.columns
group_cols = list(SubsetNode(cands_group_cols, uid="group_by_group_cols"))
new_col = FreshColumn(uid="summarise_new_col")
agg_cands = ["count", "mean", "sum"]
agg = SelectConst(agg_cands, uid="summarise_agg")
if agg != "count":
cands_agg_col = [c for c in g_df.columns if c.value not in group_cols]
agg_col = SelectNode(cands_agg_col, uid="summarise_col")
else:
cands_agg_col = None
agg_col = None
summaries = {new_col: (agg_col, agg)}
result_groupby = RInterpreter.group_by(df, group_cols)
result = RInterpreter.summarise(result_groupby, summaries)
call_str = f"summarise(group_by({{inp1}}, group_cols={group_cols!r}), summaries={{{summaries!r}}})"
# -------------------------------------------------------------------------------------------------------------- #
# Graph Construction
# -------------------------------------------------------------------------------------------------------------- #
g_res = DataFrameGraph(result)
graph = GraphRLang.assemble([g_df, g_res])
added_edges: List[Edge] = []
col_map_df = {c.value: c
for c in g_df.columns
} # Map from df's columns to their column nodes
col_map_res = {c.value: c
for c in g_res.columns
} # Map from result's columns to their column nodes
# - EQUAL edges between group columns and corresponding columns in output.
for c in group_cols:
added_edges.append(Edge(col_map_df[c], col_map_res[c], ELabel.EQUAL))
# - EQUAL edges between the cells of the group columns and the corresponding cells in the output.
# - SUM/MEAN/COUNT edges between the cells of aggregated columns (or group columns in case of 'count') and the
# resulting cells in the output.
agg_cols = [agg_col] if agg != "count" else group_cols
for idx, group in enumerate(result.loc[:, group_cols].values):
group = group[0] if len(group) == 1 else tuple(group)
df_indices = result_groupby.groups[
group] # Get the indices in df that correspond to this group
out_node = g_res.loc[result.index[idx], new_col]
interm_node = graph.create_intermediate_node(out_node.value)
added_edges.append(Edge(interm_node, out_node, ELabel.EQUAL))
# Aggregation edges
for col in agg_cols:
for val_node in g_df.loc[df_indices, col]:
added_edges.append(
Edge(val_node, interm_node, getattr(ELabel, agg.upper())))
# Equality edges for group_col nodes
for col in group_cols:
group_node = g_res.loc[result.index[idx], col]
for df_group_node in g_df.loc[df_indices, col]:
added_edges.append(
Edge(df_group_node, group_node, ELabel.EQUAL))
# - EQUAL edge between the input deletion node and the output deletion node.
added_edges.append(
Edge(g_df.deletion_node, g_res.deletion_node, ELabel.EQUAL))
# - DELETE edges between the non-group cols and non-aggregated columns and their cells
# to the deletion node of the output.
# - DELETE edge between the column of the aggregated column (in case of sum/mean) and the deletion node
# of the output.
non_group_cols = [c for c in df.columns if c not in group_cols]
for col in non_group_cols:
added_edges.append(
Edge(col_map_df[col], g_res.deletion_node, ELabel.DELETE))
if col != agg_col:
for cell in g_df.loc[:, col]:
added_edges.append(
Edge(cell, g_res.deletion_node, ELabel.DELETE))
# Add all the edges to the graph in one go.
graph.add_nodes_and_edges(edges=added_edges)
# -------------------------------------------------------------------------------------------------------------- #
# Add information about arguments
# -------------------------------------------------------------------------------------------------------------- #
tagged_edges: List[TaggedEdge] = []
for c_node in cands_group_cols:
if c_node.value in group_cols:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@group_by_group_cols"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@group_by_group_cols"))
if cands_agg_col is not None:
for c_node in cands_agg_col:
if c_node.value == agg_col:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="SELECTED@summarise_col"))
else:
tagged_edges.append(
TaggedEdge(src=c_node,
dst=c_node,
tag="NOT_SELECTED@summarise_col"))
graph.add_tags([
f"SELECTED@{cand}@summarise_agg"
if cand == agg else f"NOT_SELECTED@{cand}@summarise_agg"
for cand in agg_cands
])
graph.add_tagged_edges(tagged_edges)
return result, call_str, graph, g_res
def process_ui_interaction(
self, inputs: Dict[str, Any],
interactions: List[Dict]) -> Tuple[Any, Graph, Dict[str, Graph]]:
value_interactions = [
interaction for interaction in interactions
if interaction['to'] != ""
]
output_cells = [[int(r), int(c)]
for r, c, _ in (interaction['to'].split(':')
for interaction in value_interactions)]
if len(output_cells) > 0:
row_nums, col_nums = list(zip(*output_cells))
else:
row_nums = []
col_nums = []
min_r, max_r = min([i for i in row_nums if i >= 0] +
[0]), max([i for i in row_nums if i >= 0] + [0])
min_c, max_c = min(col_nums, default=0), max(col_nums, default=0)
num_rows = max_r - min_r + 1
num_cols = max_c - min_c + 1
output = pd.DataFrame([[f"_CELL_{r}_{c}" for c in range(num_cols)]
for r in range(num_rows)],
columns=[f"_COL_{c}" for c in range(num_cols)])
columns = list(output.columns)
for interaction in value_interactions:
value = interaction['value']
r, c, _ = interaction['to'].split(':')
r = int(r)
c = int(c)
if r == -1:
columns[c - min_c] = value
else:
output.iloc[r - min_r, c - min_c] = value
output.columns = columns
graph = GraphRLang()
g_inputs: Dict[str, DataFrameGraph] = {
key: DataFrameGraph(inp)
for key, inp in inputs.items()
}
g_output = DataFrameGraph(output)
for g_inp in g_inputs.values():
graph.merge(g_inp)
graph.merge(g_output)
for interaction in interactions:
if interaction["labels"] == ["DELETE"]:
node_to = g_output.deletion_node
for r_from, c_from, inp_id in (i.split(':')
for i in interaction['from']):
node_from = g_inputs[inp_id].get_node_xy(
int(r_from), int(c_from))
graph.add_edge(Edge(node_from, node_to, ELabel.DELETE))
continue
value = interaction['value']
r_to, c_to, _ = interaction['to'].split(':')
r_to = int(r_to)
c_to = int(c_to)
if r_to >= 0:
r_to -= min_r
c_to -= min_c
if r_to == -1:
actual_value = output.columns[c_to]
else:
actual_value = output.iloc[r_to, c_to]
if _not_equal(actual_value, value):
continue
node_to = g_output.get_node_xy(r_to, c_to)
node_from = _generate_computation_node(interaction, graph,
g_inputs)
graph.add_edge(Edge(node_from, node_to, ELabel.EQUAL))
return output, graph, g_inputs
