Replace float64 and int64 with float32 and int32, respectively

python/tests/beagle.py

@@ -87,9 +87,9 @@ def get_weights(genotyped_pos, ungenotyped_pos):
     genotyped_cm = convert_to_genetic_map_position(genotyped_pos)
     ungenotyped_cm = convert_to_genetic_map_position(ungenotyped_pos)
     # Calculate weights for ungenotyped markers.
-    weights = np.zeros(x, dtype=np.float64)
+    weights = np.zeros(x, dtype=np.float32)
     # Identify genotype markers k and k + 1 sandwiching marker i.
-    marker_interval_start = np.zeros(x, dtype=np.int64)
+    marker_interval_start = np.zeros(x, dtype=np.int32)
     for i in np.arange(x):
         if ungenotyped_pos[i] < genotyped_pos[0]:
             # Ungenotyped marker is before the first genotyped marker.
@@ -126,7 +126,7 @@ def get_mismatch_prob(pos, miscall_rate):
     assert isinstance(miscall_rate, float)
     if miscall_rate >= 0.5:
         miscall_rate = 0.5
-    mu = np.zeros(len(pos), dtype=np.float64) + miscall_rate
+    mu = np.zeros(len(pos), dtype=np.float32) + miscall_rate
     return mu
 
 
@@ -150,7 +150,7 @@ def get_switch_prob(pos, h, ne):
     assert isinstance(ne, float), "Effective population size is not a float."
     # Get genetic distance between adjacent markers.
     cm = convert_to_genetic_map_position(pos)
-    d = np.zeros(len(pos), dtype=np.float64)
+    d = np.zeros(len(pos), dtype=np.float32)
     d[0] = cm[0]
     d[1:] = cm[1:] - cm[:-1]
     assert len(d) == len(pos)
@@ -185,7 +185,7 @@ def compute_forward_probability_matrix(ref_h, query_h, rho, mu):
     assert m == len(mu)
     assert 0 <= np.min(rho) and np.max(rho) <= 1
     assert 0 <= np.min(mu) and np.max(mu) <= 0.5
-    fm = np.zeros((m, h), dtype=np.float64)
+    fm = np.zeros((m, h), dtype=np.float32)
     last_sum = 1.0  # Part of normalization factor
     for i in np.arange(m):
         # Get site-specific parameters.
@@ -236,7 +236,7 @@ def compute_backward_probability_matrix(ref_h, query_h, rho, mu):
     assert m == len(mu)
     assert 0 <= np.min(rho) and np.max(rho) <= 1
     assert 0 <= np.min(mu) and np.max(mu) <= 0.5
-    bm = np.zeros((m, h), dtype=np.float64)
+    bm = np.zeros((m, h), dtype=np.float32)
     bm[-1, :] = 1.0 / h  # Initialise the last column
     for i in np.arange(m - 2, -1, -1):
         query_a = query_h[i + 1]
@@ -330,9 +330,9 @@ def _compute_state_probability_matrix(fm, bm, ref_h, query_h, rho, mu):
     assert 0 <= np.min(rho) and np.max(rho) <= 1
     assert 0 <= np.min(mu) and np.max(mu) <= 0.5
     # m + 1 columns to allow for imputed markers right of the last genotyped marker.
-    sm = np.zeros((m + 1, h), dtype=np.float64)
-    fwd_hap_probs = np.zeros((m, 4), dtype=np.float64)
-    bwd_hap_probs = np.zeros((m, 4), dtype=np.float64)
+    sm = np.zeros((m + 1, h), dtype=np.float32)
+    fwd_hap_probs = np.zeros((m, 4), dtype=np.float32)
+    bwd_hap_probs = np.zeros((m, 4), dtype=np.float32)
     for i in np.arange(m - 1, -1, -1):
         for j in np.arange(h):
             sm[i, j] = fm[i, j] * bm[i, j]
@@ -384,7 +384,7 @@ def _interpolate_allele_probabilities(
     assert x == len(imputed_cm)
     weights, _ = get_weights(genotyped_cm, imputed_cm)
     assert x == len(weights)
-    i_hap_probs = np.zeros((x, 2), dtype=np.float64)
+    i_hap_probs = np.zeros((x, 2), dtype=np.float32)
     for i in np.arange(x):
         # Identify the genotyped markers k and k + 1 sandwiching ungenotyped marker i.
         if ungenotyped_pos[i] < genotyped_pos[0]:
@@ -426,7 +426,7 @@ def compute_state_probability_matrix(fm, bm):
     assert not np.any(bm < 0), "Backward probability matrix has negative values."
     assert not np.any(np.isnan(bm)), "Backward probability matrix has NaN values."
     # m + 1 columns to allow for imputed markers right of the last genotyped marker.
-    sm = np.zeros((m + 1, h), dtype=np.float64)
+    sm = np.zeros((m + 1, h), dtype=np.float32)
     sm[:-1, :] = np.multiply(fm, bm)
     sm[m, :] = sm[m - 1, :]  # Not in BEAGLE
     return sm
@@ -464,7 +464,7 @@ def interpolate_allele_probabilities(sm, ref_h, genotyped_pos, ungenotyped_pos):
     weights, marker_interval_start = get_weights(genotyped_pos, ungenotyped_pos)
     assert x == len(weights)
     assert x == len(marker_interval_start)
-    p = np.zeros((x, len(alleles)), dtype=np.float64)
+    p = np.zeros((x, len(alleles)), dtype=np.float32)
     # Compute allele probabilities as per Equation 1 in BB2016.
     for a in alleles:
         # Going from left to right.

python/tests/beagle_numba.py

@@ -47,9 +47,9 @@ def get_weights(genotyped_pos, ungenotyped_pos):
     genotyped_cm = convert_to_genetic_map_position(genotyped_pos)
     ungenotyped_cm = convert_to_genetic_map_position(ungenotyped_pos)
     # Calculate weights for ungenotyped markers.
-    weights = np.zeros(x, dtype=np.float64)
+    weights = np.zeros(x, dtype=np.float32)
     # Identify genotype markers k and k + 1 sandwiching ungenotyped marker i.
-    marker_interval_start = np.zeros(x, dtype=np.int64)
+    marker_interval_start = np.zeros(x, dtype=np.int32)
     idx_m = m - 1
     # Going from right to left.
     for idx_x in np.arange(x - 1, -1, -1):
@@ -86,7 +86,7 @@ def get_mismatch_prob(pos, miscall_rate):
     """
     if miscall_rate >= 0.5:
         miscall_rate = 0.5
-    mu = np.zeros(len(pos), dtype=np.float64) + miscall_rate
+    mu = np.zeros(len(pos), dtype=np.float32) + miscall_rate
     return mu
 
 
@@ -107,7 +107,7 @@ def get_switch_prob(pos, h, ne):
     """
     # Get genetic distances between adjacent markers.
     cm = convert_to_genetic_map_position(pos)
-    d = np.zeros(len(pos), dtype=np.float64)
+    d = np.zeros(len(pos), dtype=np.float32)
     d[0] = cm[0]
     d[1:] = cm[1:] - cm[:-1]
     rho = -np.expm1(-(4 * ne * d / h))
@@ -131,7 +131,7 @@ def compute_forward_probability_matrix(ref_h, query_h, rho, mu):
     """
     m = ref_h.shape[0]
     h = ref_h.shape[1]
-    fm = np.zeros((m, h), dtype=np.float64)
+    fm = np.zeros((m, h), dtype=np.float32)
     last_sum = 1.0  # Part of normalization factor
     for i in np.arange(m):
         # Get site-specific parameters
@@ -173,7 +173,7 @@ def compute_backward_probability_matrix(ref_h, query_h, rho, mu):
     """
     m = ref_h.shape[0]
     h = ref_h.shape[1]
-    bm = np.zeros((m, h), dtype=np.float64)
+    bm = np.zeros((m, h), dtype=np.float32)
     bm[-1, :] = 1.0 / h  # Initialise the last column
     for i in np.arange(m - 2, -1, -1):
         query_a = query_h[i + 1]
@@ -211,7 +211,7 @@ def compute_state_probability_matrix(fm, bm):
     m = fm.shape[0]
     h = bm.shape[1]
     # m + 1 columns to allow for imputed markers right of the last genotyped marker.
-    sm = np.zeros((m + 1, h), dtype=np.float64)
+    sm = np.zeros((m + 1, h), dtype=np.float32)
     sm[:-1, :] = np.multiply(fm, bm)
     sm[m, :] = sm[m - 1, :]  # Not in BEAGLE
     return sm
@@ -242,7 +242,7 @@ def interpolate_allele_probabilities(sm, ref_h, genotyped_pos, ungenotyped_pos):
     alleles = np.arange(4)  # ACGT encoding
     x = len(ungenotyped_pos)
     weights, marker_interval_start = get_weights(genotyped_pos, ungenotyped_pos)
-    p = np.zeros((x, len(alleles)), dtype=np.float64)
+    p = np.zeros((x, len(alleles)), dtype=np.float32)
     for a in alleles:
         for i in np.arange(x):
             # Identify genotyped markers k and k + 1 sandwiching ungenotyped marker i.