feat: implement lstsq

ivy/functional/frontends/tensorflow/linalg.py

@@ -177,6 +177,49 @@ def logdet(matrix, name=None):
     return ivy.det(matrix).log()
 
 
+@to_ivy_arrays_and_back
+@with_supported_dtypes(
+    {"2.13.0 and below": ("float32", "float64", "complex64")}, "tensorflow"
+)
+def lstsq(
+    matrix,
+    rhs,
+    l2_regularizer=0.0,
+    fast=True,
+):
+    matrix_num_dim = matrix.get_num_dims()
+    rhs_num_dim = rhs.get_num_dims()
+    if matrix_num_dim < 2:
+        raise RuntimeError("input must have at least 2 dimensions. ")
+    if matrix_num_dim - rhs_num_dim <= 1:
+        for i in range(
+            matrix_num_dim - 1
+        ):  # should have the same batch shape and same m shape
+            if matrix.shape[i] != rhs.shape[i]:
+                raise RuntimeError(f" input.size({i}) should match other.size({i})")
+    else:
+        raise RuntimeError(
+            "input.dim() must be greater or equal to other.dim() and (input.dim() -"
+            " other.dim()) <= 1"
+        )
+    if l2_regularizer != 0:
+        raise NotImplementedError(
+            "linalg.lstsq is currently disabled for complex128 and l2_regularizer != 0"
+            " due to poor accuracy."
+        )
+
+    matrix_dtype = matrix.dtype
+    matrix = ivy.astype(matrix, ivy.float64)
+    rhs = ivy.astype(rhs, ivy.float64)
+
+    q, r = ivy.qr(matrix)
+    r_inv = ivy.pinv(r)
+    solution = ivy.matmul(ivy.matmul(r_inv, ivy.matrix_transpose(q)), rhs)
+    solution = ivy.astype(solution, matrix_dtype)
+
+    return solution
+
+
 @to_ivy_arrays_and_back
 def lu_matrix_inverse(lower_upper, perm, validate_args=False, name=None):
     return ivy.lu_matrix_inverse(

ivy_tests/test_ivy/test_frontends/test_tensorflow/test_linalg.py

@@ -93,6 +93,22 @@ def _get_dtype_and_rank_2k_tensors(draw):
     )
 
 
+@st.composite
+def _get_dtype_and_same_dim_matrix(draw):
+    randam_shape = draw(helpers.get_shape(min_num_dims=2, max_num_dims=4))
+    dtype_and_values = draw(
+        helpers.dtype_and_values(
+            available_dtypes=helpers.get_dtypes(kind="valid", full=False),
+            num_arrays=2,
+            shape=randam_shape,
+            shared_dtype=True,
+            min_value=1,
+            max_value=4,
+        )
+    )
+    return dtype_and_values
+
+
 @st.composite
 def _get_dtype_and_sequence_of_arrays(draw):
     array_dtype = draw(helpers.get_dtypes("float", full=False))
@@ -747,6 +763,34 @@ def test_tensorflow_logdet(
     )
 
 
+@handle_frontend_test(
+    fn_tree="tensorflow.linalg.lstsq", dtype_x=_get_dtype_and_same_dim_matrix()
+)
+def test_tensorflow_lstsq(
+    *,
+    dtype_x,
+    on_device,
+    fn_tree,
+    frontend,
+    test_flags,
+    backend_fw,
+):
+    input_dtype, values = dtype_x
+    test_flags.num_positional_args = 2
+    helpers.test_frontend_function(
+        input_dtypes=input_dtype,
+        backend_to_test=backend_fw,
+        test_flags=test_flags,
+        frontend=frontend,
+        fn_tree=fn_tree,
+        on_device=on_device,
+        rtol=1e-03,
+        atol=1e-03,
+        a=values[0],
+        b=values[1],
+    )
+
+
 @handle_frontend_test(
     fn_tree="tensorflow.linalg.matmul",
     dtype_x=helpers.dtype_and_values(