Use scipy.linalg.lstsq instead of np.linalg.lstsq (#83)

* Use scipy.linalg.lstsq

* Remove line_profielr line.

* gelsy does not support empty target.

ivmodels/utils.py

@@ -1,4 +1,5 @@
 import numpy as np
+import scipy
 
 try:
     import pandas as pd
@@ -36,13 +37,24 @@ def proj(Z, *args):
             raise ValueError(f"Shape mismatch: Z.shape={Z.shape}, f.shape={f.shape}.")
 
     if len(args) == 1:
-        return np.dot(Z, np.linalg.lstsq(Z, args[0], rcond=None)[0])
+        # The gelsy driver raises in this case - we handle it separately
+        if len(args[0].shape) == 2 and args[0].shape[1] == 0:
+            return np.zeros_like(args[0])
+
+        return np.dot(
+            Z, scipy.linalg.lstsq(Z, args[0], cond=None, lapack_driver="gelsy")[0]
+        )
 
     csum = np.cumsum([f.shape[1] if len(f.shape) == 2 else 1 for f in args])
     csum = [0] + csum.tolist()
 
     fs = np.hstack([f.reshape(Z.shape[0], -1) for f in args])
-    fs = np.dot(Z, np.linalg.lstsq(Z, fs, rcond=None)[0])
+
+    if fs.shape[1] == 0:
+        # The gelsy driver raises in this case - we handle it separately
+        return (*(np.zeros_like(f) for f in args),)
+
+    fs = np.dot(Z, scipy.linalg.lstsq(Z, fs, cond=None, lapack_driver="gelsy")[0])
 
     return (
         *(fs[:, i:j].reshape(f.shape) for i, j, f in zip(csum[:-1], csum[1:], args)),
@@ -68,26 +80,11 @@ def oproj(Z, *args):
     if Z is None:
         return (*args,)
 
-    for f in args:
-        if len(f.shape) > 2:
-            raise ValueError(
-                f"*args should have shapes (n, d_f) or (n,). Got {f.shape}."
-            )
-        if f.shape[0] != Z.shape[0]:
-            raise ValueError(f"Shape mismatch: Z.shape={Z.shape}, f.shape={f.shape}.")
-
     if len(args) == 1:
-        return args[0] - np.dot(Z, np.linalg.lstsq(Z, args[0], rcond=None)[0])
-
-    csum = np.cumsum([f.shape[1] if len(f.shape) == 2 else 1 for f in args])
-    csum = [0] + csum.tolist()
-
-    fs = np.hstack([f.reshape(Z.shape[0], -1) for f in args])
-    fs = fs - np.dot(Z, np.linalg.lstsq(Z, fs, rcond=None)[0])
+        return args[0] - proj(Z, args[0])
 
-    return (
-        *(fs[:, i:j].reshape(f.shape) for i, j, f in zip(csum[:-1], csum[1:], args)),
-    )
+    else:
+        return (*(x - x_proj for x, x_proj in zip(args, proj(Z, *args))),)
 
 
 def to_numpy(x):

tests/test_utils.py

@@ -36,6 +36,11 @@ def test_proj_multiple_args():
     )
     assert np.allclose(proj(X, z1), X @ np.linalg.inv(X.T @ X) @ X.T @ z1)
     assert np.allclose(proj(X, z2), X @ np.linalg.inv(X.T @ X) @ X.T @ z2)
+    assert np.allclose(
+        proj(X, z2, z2),
+        X @ np.linalg.inv(X.T @ X) @ X.T @ z2,
+        X @ np.linalg.inv(X.T @ X) @ X.T @ z2,
+    )
     assert np.allclose(proj(X, z3), X @ np.linalg.inv(X.T @ X) @ X.T @ z3)