treewide: Don't use numpy asfarray (#3158)

Drop redundant uses.
Switch to 'asarray' for cases that provide explicit type.
Switch to 'asarray(..., dtype=float)' for all other cases.

Avoid type overflow in seed calculation.

psyneulink/core/components/component.py

@@ -1425,7 +1425,7 @@ def _get_state_initializer(self, context):
         def _convert(p):
             x = p.get(context)
             if p.name == 'matrix': # Flatten matrix
-                val = tuple(np.asfarray(x).flatten())
+                val = tuple(np.asarray(x, dtype=float).ravel())
             elif isinstance(x, np.random.RandomState):
                 state = x.get_state(legacy=False)
 
@@ -1626,7 +1626,7 @@ def _get_values(p):
             elif p.name == 'num_trials_per_estimate': # Should always be int
                 return 0 if param is None else int(param)
             elif p.name == 'matrix': # Flatten matrix
-                return tuple(np.asfarray(param).flatten())
+                return tuple(np.asarray(param, dtype=float).ravel())
             return _convert(param)
 
         return tuple(map(_get_values, self._get_compilation_params()))

psyneulink/core/components/functions/function.py

@@ -835,7 +835,7 @@ def convert_output_type(self, value, output_type=None):
         # Note: if 2D or 1D array has more than two items, generate exception
         elif output_type is FunctionOutputType.NP_0D_ARRAY:
             if object_has_single_value(value):
-                value = np.asfarray(value)
+                value = np.asarray(value, dtype=float)
             else:
                 raise FunctionError(f"Can't convert value ({value}) with more than a single number to a raw number.")
 

psyneulink/core/components/functions/stateful/memoryfunctions.py

@@ -1289,10 +1289,10 @@ def _parse_selection_function_variable(self, variable, context=None, distance_re
         distance_result = self.distance_function(self._parse_distance_function_variable(variable), context=context)
         # TEST PRINT:
         # print(distance_result, self.distance_function.defaults.value)
-        return np.asfarray([
+        return np.asarray([
             distance_result if i == 0 else np.zeros_like(distance_result)
             for i in range(self.defaults.max_entries)
-        ])
+        ], dtype=float)
 
     def _validate(self, context=None):
         """Validate distance_function, selection_function and memory store"""
@@ -1328,10 +1328,10 @@ def _validate(self, context=None):
 
             # Default to full memory
             selection_function = self.selection_function
-            test_var = np.asfarray([
+            test_var = np.asarray([
                 distance_result if i == 0 else np.zeros_like(distance_result)
                 for i in range(self._get_current_parameter_value('max_entries', context))
-            ])
+            ], dtype=float)
             try:
                 result = np.asarray(selection_function(test_var, context=context))
             except Exception as e:
@@ -2313,10 +2313,10 @@ def _parse_selection_function_variable(self, variable, context=None):
         distance_result = self.distance_function.parameters.value._get(context)
         # TEST PRINT:
         # print(distance_result, self.distance_function.defaults.value)
-        return np.asfarray([
+        return np.asarray([
             distance_result if i == 0 else np.zeros_like(distance_result)
             for i in range(self.defaults.max_entries)
-        ])
+        ], dtype=float)
 
     def _get_state_ids(self):
         return super()._get_state_ids() + ["ring_memory"]
@@ -2553,9 +2553,10 @@ def _validate(self, context=None):
 
         # Default to full memory
         selection_function = self.selection_function
-        test_var = np.asfarray([distance_result if i==0
-                                else np.zeros_like(distance_result)
-                                for i in range(self._get_current_parameter_value('max_entries', context))])
+        test_var = np.asarray([distance_result if i==0 else np.zeros_like(distance_result)
+                               for i in range(self._get_current_parameter_value('max_entries', context))],
+                              dtype=float)
+
         if isinstance(selection_function, type):
             selection_function = selection_function(default_variable=test_var, context=context)
             fct_string = 'Function type'
@@ -2709,12 +2710,12 @@ def _function(self,
         # Store variable to dict:
         rate = self._get_current_parameter_value(RATE, context)
         if rate is not None:
-            key = np.asfarray(key) * np.asfarray(rate)
+            key = np.asarray(key, dtype=float) * np.asarray(rate, dtype=float)
             assert len(key) == len(variable[KEYS]), "{} vs. {}".format(key, variable[KEYS])
 
         if noise is not None:
             # TODO: does val need noise?
-            key = np.asfarray(key) + np.asfarray(noise)[KEYS]
+            key = np.asarray(key, dtype=float) + np.asarray(noise, dtype=float)[KEYS]
             assert len(key) == len(variable[KEYS]), "{} vs. {}".format(key, variable[KEYS])
 
         if storage_prob == 1.0 or (storage_prob > 0.0 and storage_prob > random_state.uniform()):

psyneulink/core/globals/utilities.py

@@ -1707,12 +1707,12 @@ def seed(self, seed):
         assert False, "Use 'seed' parameter instead of seeding the random state directly"
 
 
-_seed = np.int32((time.time() * 1000) % 2**31)
+_seed = np.uint32((time.time() * 1000) % 2**31)
 def get_global_seed(offset=1):
     global _seed
-    _seed += offset
-    _seed %= 2**31
-    return _seed - offset
+    old_seed = _seed
+    _seed = (_seed + offset) % 2**31
+    return old_seed
 
 
 def set_global_seed(new_seed):

psyneulink/core/llvm/builder_context.py

@@ -485,7 +485,7 @@ def _param_struct(p):
             if isinstance(val, ContentAddressableList):
                 return ir.LiteralStructType(self.get_param_struct_type(x) for x in val)
             elif p.name == 'matrix':   # Flatten matrix
-                val = np.asfarray(val).flatten()
+                val = np.asarray(val, dtype=float).ravel()
             elif p.name == 'num_trials_per_estimate':  # Should always be int
                 val = np.int32(0) if val is None else np.int32(val)
             elif np.ndim(val) == 0 and component._is_param_modulated(p):
@@ -508,7 +508,7 @@ def _state_struct(p):
             if isinstance(val, ContentAddressableList):
                 return ir.LiteralStructType(self.get_state_struct_type(x) for x in val)
             if p.name == 'matrix':   # Flatten matrix
-                val = np.asfarray(val).flatten()
+                val = np.asarray(val, dtype=float).ravel()
             struct = self.convert_python_struct_to_llvm_ir(val)
             return ir.ArrayType(struct, p.history_min_length + 1)
 

psyneulink/core/llvm/execution.py

@@ -262,7 +262,7 @@ def _state_struct(self):
         return self._get_compilation_param('_state', '_get_state_initializer', 1)
 
     def execute(self, variable):
-        new_variable = np.asfarray(variable, dtype=self._bin_func.np_arg_dtypes[2].base)
+        new_variable = np.asarray(variable, dtype=self._bin_func.np_arg_dtypes[2].base)
         data_in = new_variable.reshape(self._bin_func.np_arg_dtypes[2].shape)
 
         data_out = self._bin_func.np_buffer_for_arg(3)
@@ -273,7 +273,7 @@ def execute(self, variable):
 
     def cuda_execute(self, variable):
         # Create input argument, PyCUDA doesn't care about shape
-        data_in = np.asfarray(variable, dtype=self._bin_func.np_arg_dtypes[2].base)
+        data_in = np.asarray(variable, dtype=self._bin_func.np_arg_dtypes[2].base)
         data_out = self._bin_func.np_buffer_for_arg(3)
 
         self._bin_func.cuda_call(self._cuda_param_struct,

tests/composition/pec/test_stab_flex_pec_fit.py

@@ -1,4 +1,5 @@
 import psyneulink as pnl
+import pytest
 
 import optuna
 
@@ -384,6 +385,7 @@ def run_stab_flex_cond(
 
     return comp, df
 
+@pytest.mark.llvm
 def test_stab_flex_cond_fit():
     from psyneulink.core.globals.utilities import set_global_seed
 

tests/composition/test_composition.py

@@ -7267,8 +7267,8 @@ def test_save_state_before_simulations(self):
                         C.parameters.value.get(comp)[0]]
 
         np.testing.assert_allclose(run_2_values, run_3_values)
-        np.testing.assert_allclose(np.asfarray(run_1_values), [[0.36], [0.056], [0.056]])
-        np.testing.assert_allclose(np.asfarray(run_2_values), [[0.5904], [0.16384], [0.16384]])
+        np.testing.assert_allclose(run_1_values, [[0.36], [0.056], [0.056]])
+        np.testing.assert_allclose(run_2_values, [[0.5904], [0.16384], [0.16384]])
 
     def test_reset_clear_results(self):
         mech = ProcessingMechanism(name='mech')

tests/composition/test_control.py

@@ -3354,7 +3354,7 @@ def test_model_based_ocm(self, benchmark, controller_mode, result, mode, ocm_mod
 
         def comp_run(inputs, execution_mode):
             comp.run(inputs=inputs, execution_mode=execution_mode)
-            return comp.results.copy(), np.asfarray(ocm.function.saved_values)
+            return comp.results.copy(), np.asarray(ocm.function.saved_values)
 
         results, saved_values = benchmark(comp_run, inputs, mode)
 

tests/functions/test_buffer.py

@@ -88,8 +88,9 @@ def test_buffer_initializer_len_3(self, benchmark):
                    history=3)
         np.testing.assert_allclose(B.execute(3.0), [[1.0], [2.0], np.array([3.])])
         np.testing.assert_allclose(B.execute(4.0), [[2.0], np.array([3.]), np.array([4.])])
+
         val = benchmark(B.execute, 5.0)
-        np.testing.assert_allclose(val, [np.array([3.]), np.array([4.]), np.array([5.])])
+        np.testing.assert_array_equal(val, [[3.], [4.], [5.]])
 
     @pytest.mark.benchmark(group="BufferFunction")
     def test_buffer_as_function_of_processing_mech(self, benchmark):
@@ -100,7 +101,7 @@ def test_buffer_as_function_of_processing_mech(self, benchmark):
         val = benchmark(P.execute, 1.0)
 
         # NOTE: actual output is [0, [[1]]]
-        np.testing.assert_allclose(np.asfarray(val, dtype=object), [[0., 1.]])
+        np.testing.assert_allclose(val, [[0., 1.]])
 
         # fails due to value and variable problems when Buffer is the function of a mechanism
         # P = ProcessingMechanism(function=Buffer(default_variable=[[0.0], [1.0], [2.0]],
@@ -121,17 +122,15 @@ def test_buffer_as_function_of_origin_mech_in_composition(self):
         def assemble_full_result():
             full_result.append(P.parameters.value.get(C))
 
-        C.run(inputs={P: [[1.0], [2.0], [3.0], [4.0], [5.0]]},
-              call_after_trial=assemble_full_result)
+        C.run(inputs={P: [[1.0], [2.0], [3.0], [4.0], [5.0]]}, call_after_trial=assemble_full_result)
         # only returns index 0 item of the deque on each trial  (OutputPort value)
-        np.testing.assert_allclose(np.asfarray(C.results), [[[0.0]], [[0.0]], [[1.0]], [[2.0]], [[3.0]]])
+        np.testing.assert_allclose(C.results, [[[0.0]], [[0.0]], [[1.0]], [[2.0]], [[3.0]]])
 
         # stores full mechanism value (full deque) on each trial
-        expected_full_result = [np.array([[0.], [1.]]),
-                                np.array([[0.], [1.], [2.]]),
-                                np.array([[1.], [2.], [3.]]),   # Shape change
-                                np.array([[2.], [3.], [4.]]),
-                                np.array([[3.], [4.], [5.]])]
+        expected_full_result = [[[0.], [1.]],
+                                [[0.], [1.], [2.]],
+                                [[1.], [2.], [3.]],   # Shape change
+                                [[2.], [3.], [4.]],
+                                [[3.], [4.], [5.]]]
         for i in range(5):
-            np.testing.assert_allclose(expected_full_result[i],
-                               np.asfarray(full_result[i]))
+            np.testing.assert_array_equal(expected_full_result[i], full_result[i])

tests/llvm/test_helpers.py

@@ -42,12 +42,12 @@ def test_helper_fclamp(mode):
 
         builder.ret_void()
 
-    ref = np.clip(VECTOR, TST_MIN, TST_MAX)
-    bounds = np.asfarray([TST_MIN, TST_MAX])
-
     bin_f = pnlvm.LLVMBinaryFunction.get(custom_name, ctype_ptr_args=(0, 2))
     local_vec = VECTOR.copy()
 
+    ref = np.clip(VECTOR, TST_MIN, TST_MAX)
+    bounds = np.asarray([TST_MIN, TST_MAX], dtype=bin_f.np_arg_dtypes[2].base)
+
     if mode == 'CPU':
         ct_vec = local_vec.ctypes.data_as(bin_f.c_func.argtypes[0])
         ct_bounds = bounds.ctypes.data_as(bin_f.c_func.argtypes[2])
@@ -141,8 +141,8 @@ def test_helper_is_close(mode, var1, var2, rtol, atol, fp_type):
 
     bin_f = pnlvm.LLVMBinaryFunction.get(custom_name, ctype_ptr_args=(0, 1, 2))
 
-    vec1 = np.atleast_1d(np.asfarray(var1, dtype=bin_f.np_arg_dtypes[0].base))
-    vec2 = np.atleast_1d(np.asfarray(var2, dtype=bin_f.np_arg_dtypes[1].base))
+    vec1 = np.atleast_1d(np.asarray(var1, dtype=bin_f.np_arg_dtypes[0].base))
+    vec2 = np.atleast_1d(np.asarray(var2, dtype=bin_f.np_arg_dtypes[1].base))
     assert len(vec1) == len(vec2)
     res = np.empty_like(vec2)
 
@@ -163,8 +163,8 @@ def test_helper_is_close(mode, var1, var2, rtol, atol, fp_type):
 @pytest.mark.parametrize('mode', ['CPU', pytest.helpers.cuda_param('PTX')])
 @pytest.mark.parametrize('rtol,atol', [[0, 0], [None, None], [None, 100], [2, None]])
 @pytest.mark.parametrize('var1,var2',
-                         [[1, 1], [1, 100], [1,2], [-4,5], [0, -100], [-1,-2],
-                          [[1,1,1,-4,0,-1], [1,100,2,5,-100,-2]]
+                         [[1, 1], [1, 100], [1, 2], [-4, 5], [0, -100], [-1, -2],
+                          [[1, 1, 1, -4, 0, -1], [1, 100, 2, 5, -100, -2]]
                          ])
 def test_helper_all_close(mode, var1, var2, atol, rtol):
 
@@ -174,8 +174,8 @@ def test_helper_all_close(mode, var1, var2, atol, rtol):
     if atol is not None:
         tolerance['atol'] = atol
 
-    vec1 = np.atleast_1d(np.asfarray(var1))
-    vec2 = np.atleast_1d(np.asfarray(var2))
+    vec1 = np.atleast_1d(var1)
+    vec2 = np.atleast_1d(var2)
     assert len(vec1) == len(vec2)
 
     with pnlvm.LLVMBuilderContext.get_current() as ctx:
@@ -194,6 +194,9 @@ def test_helper_all_close(mode, var1, var2, atol, rtol):
         builder.ret_void()
 
     bin_f = pnlvm.LLVMBinaryFunction.get(custom_name)
+    vec1 = np.atleast_1d(np.asarray(var1, dtype=bin_f.np_arg_dtypes[0].base))
+    vec2 = np.atleast_1d(np.asarray(var2, dtype=bin_f.np_arg_dtypes[1].base))
+    assert len(vec1) == len(vec2)
     res = bin_f.np_buffer_for_arg(2)
 
     ref = np.allclose(vec1, vec2, **tolerance)
@@ -438,7 +441,7 @@ def test_helper_numerical(mode, op, var, expected, fp_type):
 
     bin_f = pnlvm.LLVMBinaryFunction.get(custom_name)
 
-    res = np.asfarray(var, dtype=bin_f.np_arg_dtypes[0])
+    res = np.asarray(var, dtype=bin_f.np_arg_dtypes[0])
 
     if mode == 'CPU':
         bin_f(res)
@@ -478,8 +481,8 @@ def test_helper_recursive_iterate_arrays(mode, var1, var2, expected):
 
     bin_f = pnlvm.LLVMBinaryFunction.get(custom_name)
 
-    vec1 = np.asfarray(var1, dtype=bin_f.np_arg_dtypes[0].base)
-    vec2 = np.asfarray(var2, dtype=bin_f.np_arg_dtypes[1].base)
+    vec1 = np.asarray(var1, dtype=bin_f.np_arg_dtypes[0].base)
+    vec2 = np.asarray(var2, dtype=bin_f.np_arg_dtypes[1].base)
     res = bin_f.np_buffer_for_arg(2)
 
     if mode == 'CPU':
@@ -518,7 +521,7 @@ def test_helper_convert_fp_type(t1, t2, mode, val):
     np_dt1, np_dt2 = (np.dtype(bin_f.np_arg_dtypes[i]) for i in (0, 1))
 
     # instantiate value, result and reference
-    x = np.asfarray(val, dtype=np_dt1)
+    x = np.asarray(val, dtype=np_dt1)
     y = bin_f.np_buffer_for_arg(1)
     ref = x.astype(np_dt2)
 

tests/mechanisms/test_control_mechanism.py

@@ -187,7 +187,7 @@ def test_identicalness_of_control_and_gating(self):
         # c.add_linear_processing_pathway(pathway=z)
         comp.add_node(Control_Mechanism)
 
-        np.testing.assert_allclose(np.asfarray(Control_Mechanism.parameters.control_allocation.get()), [[0], [0], [0]])
+        np.testing.assert_allclose(Control_Mechanism.parameters.control_allocation.get(), [[0], [0], [0]])
 
         stim_list = {
             Input_Layer: [[-1, 30]],

tests/mechanisms/test_ddm_mechanism.py

@@ -29,7 +29,7 @@ def test_valid(self):
         #  returns previous_value + rate * variable * time_step_size  + noise
         #  0.0 + 1.0 * 1.0 * 1.0 + 0.0
         D.execute(1.0)
-        np.testing.assert_allclose(np.asfarray(D.value),  [[1.0], [1.0]])
+        np.testing.assert_array_equal(D.value,  [[1.0], [1.0]])
         np.testing.assert_allclose(D.output_ports[0].value[0], 1.0)
         np.testing.assert_allclose(D.output_ports[1].value[0], 1.0)
 
@@ -38,7 +38,7 @@ def test_valid(self):
         np.testing.assert_allclose(D.function.value[0], 2.0)
         np.testing.assert_allclose(D.function.parameters.previous_value.get(), 2.0)
         np.testing.assert_allclose(D.function.previous_time, 0.1)
-        np.testing.assert_allclose(np.asfarray(D.value),  [[1.0], [1.0]])
+        np.testing.assert_array_equal(D.value,  [[1.0], [1.0]])
         np.testing.assert_allclose(D.output_ports[0].value[0], 1.0)
         np.testing.assert_allclose(D.output_ports[1].value[0], 1.0)
 
@@ -47,7 +47,7 @@ def test_valid(self):
         np.testing.assert_allclose(D.function.value[0], 0.0)
         np.testing.assert_allclose(D.function.parameters.previous_value.get(), 0.0)
         np.testing.assert_allclose(D.function.previous_time, 0.0)
-        np.testing.assert_allclose(np.asfarray(D.value), [[1.0], [1.0]])
+        np.testing.assert_array_equal(D.value, [[1.0], [1.0]])
         np.testing.assert_allclose(D.output_ports[0].value[0], 1.0)
         np.testing.assert_allclose(D.output_ports[1].value[0], 1.0)
 
@@ -56,13 +56,13 @@ def test_valid(self):
         np.testing.assert_allclose(D.function.value[0], 2.0)
         np.testing.assert_allclose(D.function.parameters.previous_value.get(), 2.0)
         np.testing.assert_allclose(D.function.previous_time, 0.1)
-        np.testing.assert_allclose(np.asfarray(D.value), [[2.0], [0.1]])
+        np.testing.assert_array_equal(D.value, [[2.0], [0.1]])
         np.testing.assert_allclose(D.output_ports[0].value, 2.0)
         np.testing.assert_allclose(D.output_ports[1].value, 0.1)
 
         D.execute(1.0)
         #  2.0 + 1.0 = 3.0 ; 0.1 + 1.0 = 1.1
-        np.testing.assert_allclose(np.asfarray(D.value), [[3.0], [1.1]])
+        np.testing.assert_array_equal(D.value, [[3.0], [1.1]])
         np.testing.assert_allclose(D.output_ports[0].value[0], 3.0)
         np.testing.assert_allclose(D.output_ports[1].value[0], 1.1)
 
@@ -639,10 +639,7 @@ def test_DDM_in_composition(benchmark, comp_mode):
     inputs = {M: [10]}
     val = benchmark(C.run, inputs, num_trials=2, execution_mode=comp_mode)
 
-    # FIXME: Python version returns dtype=object
-    val = np.asfarray(val)
-    np.testing.assert_allclose(val[0], [2.0])
-    np.testing.assert_allclose(val[1], [0.2])
+    np.testing.assert_allclose(val, [[2.0], [0.2]])
 
 
 @pytest.mark.composition
@@ -812,8 +809,8 @@ def test_DDMMechanism_LCA_equivalent(comp_mode):
     ddm = DDM(default_variable=[0],
               function=DriftDiffusionIntegrator(rate=1, time_step_size=0.1),
               execute_until_finished=False)
-    comp2 = Composition()
-    comp2.add_node(ddm)
-    result2 = comp2.run(inputs={ddm:[1]}, execution_mode=comp_mode)
-    np.testing.assert_allclose(np.asfarray(result2[0]), [0.1])
-    np.testing.assert_allclose(np.asfarray(result2[1]), [0.1])
+    comp = Composition()
+    comp.add_node(ddm)
+    result = comp.run(inputs={ddm:[1]}, execution_mode=comp_mode)
+
+    np.testing.assert_allclose(result, [[0.1], [0.1]])